A Markedness Subhierarchy in Syntax - ROAroa.rutgers.edu/files/80-0000/roa-80-bakovic-2.pdf · A...

ROA version August 1995 Comments welcome!

(Revised version to appear in the proceedings of the MIT Optimality in Syntax Workshop)

A Markedness Subhierarchy in SyntaxOptimality and Inversion in Spanish*

Eric J. BakovicRutgers University

0 Introduction

Spanish exhibits a case of what has been referred to by Rizzi (1991) as residual verb second: verbsecond-like facts in a certain subset of the syntactic constructions of the language. The subset ofsyntactic constructions showing verb second in Spanish is the set of wh-questions. Following thegenerative literature on the subject (beginning with Torrego 1984 for Spanish), I will henceforthrefer to this phenomenon as inversion,1 a descriptive term highlighting the fact that when anonsubject wh-phrase is fronted in its clause, the residual verb second effect gives the invertedorder verb–subject as opposed to the normal (i.e., base-generated) order subject–verb. There issome very interesting variation to the gross generalization that Spanish exhibits inversion in wh-questions. In short, whether or not a wh-question will require inversion in a given dialect2depends on two factors: whether it is a matrix or a subordinate wh-question, and how argumentalthe moved wh-phrase is.

This latter factor was first discovered by Torrego (1984), who noted that in her dialect,argument wh-phrases (quién ‘who’ and qué ‘what’) require inversion while adjunct wh-phrases(dónde ‘where’, cuándo ‘when’, cómo ‘how’ and por qué ‘why’) do not. Goodall (1991ab) andSuñer (1994) have noted that other dialects place the wedge between wh-phrase types (if any) inplaces other than the argument/adjunct juncture. The best way to illustrate this is with a few keydialectal examples, some (1a-c) from the works cited above and the others (1d-f) from a judgmentsurvey of 30 speakers that I conducted in early 1995.

(1) Examples of dialectal variation with respect to inversion

a. No inversion with any wh-phrases. (Suñer 1994)b. Inversion with argument wh-phrases only. (Torrego 1984, Suñer 1994)c. Inversion with all but reason wh-phrases (i.e., por qué ‘why’). (Goodall 1991ab)d. Inversion with all wh-phrases in matrix clauses; all but reason wh-phrases in

subordinate clauses. (Survey results)e. Inversion with all but reason wh-phrases in matrix clauses; only argument wh-

phrases in subordinate clauses. (Survey results)f. Inversion with argument wh-phrases in matrix clauses; no inversion in subordinate

clauses. (Survey results)

* I thank my advisor Jane Grimshaw, and my committee members Veneeta Dayal and Viviane Déprez, for their invaluableassistance; Michelle Carroll, Ed Keer, Alan Prince, Vieri Samek-Lodovici, Paul Smolensky and Bruce Tesar for extremelyhelpful discussions; the members of the 1994–1995 Graduate Seminar at Rutgers for comments on various incarnations ofthis project; the audiences at MIT and at the Rutgers–UMass Joint Class Meeting for very useful feedback; and, last but notleast, my Spanish informants (family members and friends, Rutgers students, and people on the Internet) whose judgmentswere so crucial to the claims made in this paper. I claim sole responsibility for any and all errors herein.1 Torrego (1984) distinguishes between obligatory and free inversion; I concentrate on the former and call it simplyinversion, analyzing the latter differently in §4. The reader is asked to keep in mind that when I say inversion, I meanTorrego’s obligatory inversion (see fn. 16 for more on this point).2 I use the term ‘dialect’ very broadly, to refer to any variety of Spanish that is consistent with the particular patterns ofinversion to be discussed in detail below. There is no direct claim made about any other properties these varieties mayshare, nor about their geographical clustering. See §4 for more on this point.

A Markedness Subhierarchy in Syntax

2

As far as I can tell from the cited works, the dialects in (1a-c) do not show a distinction betweenmatrix and subordinate wh-questions; in any case, I have also found in my survey examples ofdialects (1a) and (1c) (but not (1b); see fn. 30). These examples show part of a more intricateimplicational relationship among wh-phrase types: inversion is either not possible (1a), requiredonly with argument (the most argumental) wh-phrases (1b), or required with all but reason (theleast argumental) wh-phrases (1c).

The diagram in (2) fills in the rest of the argumental continuum, incorporating more resultsfrom my survey. As shown, there are five patterns with respect to inversion based solely on theargumental nature of the wh-phrase. They are numbered 0–4 for convenience: inversion isrequired in the matrix/subordinate clause of dialects with pattern X with a proper subset of the wh-phrase types that require inversion in the matrix/subordinate clause of dialects with pattern X+1.

(2) Implicational relationship among wh-phrase types

A matrix/subordinate wh-question requires inversion if the wh-phrase is

1(Torrego 1984)

dónde, cuándo cómo por quéquién, qué‘how’ ‘why’

4(Survey results)

2(Survey results)

3(Goodall 1991ab)

‘who, what’ ‘where, when’

ARGUMENT LOCATION MANNER REASON

0 = no inversion (Suñer 1994)

The dialects in (1d-f), with a matrix/subordinate distinction, are exclusively from my survey.What these examples show is a subset relationship between matrix and subordinate clause types:no dialect can require inversion with more wh-phrase types in a subordinate clause than in a matrixclause. Put another way:

(3) Subset relationship between matrix and subordinate clause types

In a given dialect, subordinate clauses display inversion with a subset of the wh-phrase types that matrix clauses display inversion with.3

What this means is that out of twenty-five conceivable combinations of matrix/subordinateinversion possibilities based on the implicational relationship (2) alone (five matrix times fivesubordinate), only fifteen are actual combinations, as shown by the table in (4). Following theconvention adopted in (2), a number from 0 to 4 in a cell in the “Matrix” and “Subordinate” rowsof the table indicates which wh-phrase types require inversion in wh-questions of that clause-type(0 = none, 1 = arguments, … , 4 = all). The twenty-five possible combinations arecorrespondingly labelled with the nth letter of the alphabet, and the actual combinations that obeythe subset relationship (3) are boxed.

(4) Conceivable dialectal possibilities (actualities boxed; obeying (3))

Dialects A B C D E F G H I J K L M N O P Q R S T U V W X Y

Matrix 0 0 0 0 0 1 1 1 1 1 2 2 2 2 2 3 3 3 3 3 4 4 4 4 4

Subordinate 0 1 2 3 4 0 1 2 3 4 0 1 2 3 4 0 1 2 3 4 0 1 2 3 4

3 Note that this is not a proper subset relationship, since there are dialects with no matrix/subordinate distinction.

Eric J. Bakovic

3

In this paper, I give a complete formal account of these facts within the framework of OptimalityTheory (OT; Prince & Smolensky 1993) as applied to syntax by Grimshaw (1995). I maintain thatinversion is caused by obligatory movement of the verb to the left of the subject, into the headposition of the projection whose specifier is occupied by the wh-element. This is shownschematically in (5a); all projections are labelled VP since they are assumed to be functionalprojections in the extended projection (Grimshaw 1991) of the lexical verb. Central to the accountis my claim that wh-questions without inversion involve adjunction of the wh-element, as in (5b).4Since adjunction creates no new head position, inversion is unnecessary and impossible.

(5) Tree and bracketed diagrams of inversion vs. no inversion

a. Inversion b. Adjunction (no inversion)VP

wh V´

V VP

Subj tv twh

VP

wh VP

Subj

V twh

V´

[VP wh [V´ V [VP Subj tv twh ] ] ] [VP wh [VP Subj [V´ V twh ] ] ]

Departing slightly from Grimshaw (1995),5 I assume that wh-phrases are forced to move to a c-commanding position from which their scope is interpreted (matrix or subordinate clause-initialposition) by OP-SCOPE (6a), while they prefer to move to a clause-initial specifier position by OP-SPEC (6b). Assuming that OP-SCOPE is an inviolable principle of interpretation, then the choicebetween inversion (5a) and adjunction (5b) depends on the ranking between OP-SPEC and STAY(6c), the constraint against movement.

(6) Core principle and constraints

a. OP-SCOPE (principle of interpretation; inviolable)Syntactic operators must c-command the extended projection over which their scopeis interpreted (matrix clause if it’s a direct question, subordinate if indirect).

b. OP-SPEC (violable constraint; Grimshaw 1995)Syntactic operators must be in specifier position.

c. STAY (violable constraint; Grimshaw 1995)Trace is not allowed.

OP-SCOPE simply requires that syntactic operators have syntactic scope, regardless of the precisesyntactic position from which the scope is taken; OP-SPEC prefers said scope to be taken from aspecifier position. Since wh-phrases are syntactic operators, OP-SCOPE and OP-SPEC requiremovement of wh-phrases that aren’t base-generated in appropriately c-commanding specifierpositions. Such movement (in fact, any movement) violates STAY.

If OP-SPEC dominates STAY, then inversion will be preferred to adjunction since the wh-phrase will be in a specifier position, regardless of the fact that inversion incurs two STAYviolations (one for wh-movement, one for head movement) and adjunction incurs only one (for

4 I thank Viviane Déprez for planting the seed of this aspect of the analysis in my head. Déprez (1989, 1991) also arguesfor adjunction and movement to specifier (substitution) of wh, in a slightly different empirical and theoretical context.5 Apart from the separation of OP-SCOPE and OP-SPEC below, I propose that Grimshaw’s constraint OB-HD (OBligatoryHeaDs) is not a violable constraint. See §1.2 below for the details of this proposal.


4

wh-movement, required in any case by inviolable OP-SCOPE). If STAY dominates OP-SPEC, thenadjunction will be preferred since the extra head movement involved in inversion is not motivatedby any constraint that dominates STAY.

In §2, I discuss the issue of the “argumental nature” of wh-phrases. I propose that thedifferent types of wh-phrases are called on by the different OP-SPEC constraints in (7a-d), whichare universally ranked with respect to each other as in (7e) in what is called a markednesssubhierarchy. The implicational relationship (2) is explained by the five different ways that STAYcan be parochially ranked within the markedness subhierarchy.

(7) OP-SPEC constraint family

a. ARGOP-SPEC

Argument operators must be in specifierposition.

b. LOCOP-SPEC

Location operators must be in specifierposition.

c. MANOP-SPEC

Manner operators must be in specifierposition.

d. REASOP-SPEC

Reason operators must be in specifierposition.

e. Markedness subhierarchyARGOP-SPEC » LOCOP-SPEC » MANOP-SPEC » REASOP-SPEC

The subset relationship (3) is explained in §3 by the thirty different ways that STAY and thesubordinate clause-specific constraint PROJ-PRIN(H) can be parochially ranked within themarkedness subhierarchy.

(8) Constraint regulating subordinate clauses

PROJ-PRIN(H)No movement into the head of a subordinate clause.

Since STAY is more stringent than PROJ-PRIN(H) (a violation of the latter entails a violation of theformer but not vice versa), PROJ-PRIN(H) only has an independent effect when it is cruciallyranked above a constraint that is itself crucially ranked above STAY.6 The grammars predicted bythe thirty different ways that STAY and PROJ-PRIN(H) can be ranked within the markednesssubhierarchy are thus exactly the fifteen that obey the subset relationship (3).

Because there is only one unique output to any input in an OT grammar (save for thepossibility of output candidates that tie on the constraint hierarchy), this account predicts that whenadjunction (5b) is possible for a given wh-phrase in a given clause type in a given dialect,inversion (5a) for that wh-phrase in that clause type in that dialect is impossible, and vice versa.An account of the fact that the particular word order exemplified by inversion is possible in everydialect (Torrego’s free inversion; see fn. 1) is given in §4, which also concludes the paper with aspeculation as to why Spanish varies to the degree that it does with respect to inversion.

The account proposed here naturally explains the range of dialectal variation that exists inSpanish, solely through the interaction of conflicting constraints that are ranked and violable. Thistype of constraint interaction is the defining property of OT. The concept of a markednesshierarchy such as the one proposed here finds precedents in the now vast amount of work in OT:from Prince & Smolensky (1993), who propose one to explain the sensitivity of syllabification to

6 Ranking PROJ-PRIN(H) just above STAY is no different than ranking it anywhere below STAY, because the two constraintsdo not themselves directly conflict. See §3 below for details.

Eric J. Bakovic

5

the sonority hierarchy in phonology; to Legendre et al. (1995), who propose one to explain thesensitivity of long-distance wh-movement to the number of barriers crossed.

One of the main theoretical goals of this paper is to provide a case that highlights howmarkedness subhierarchies within OT directly explain implicational relations such as: “if anobstruent can be a coda, so can a nasal;” “if crossing one barrier is bad, so is crossing two;” “ifinversion is required with a reason wh-phrase, it is also required with an argument wh-phrase.”The independent rankings of conflicting constraints within a markedness subhierarchy, and theforced violation of constraints by dominating and conflicting (crucially dominating) constraints,explain implicational relations such as these in a straightforward manner. This effectively placesthe burden of grammatical variation not on the lexicon and in the form of abstract properties ofalready abstract categories, but on the grammar and in the form of different rankings of verygeneral, substantive constraints, all of which are universally present, many of which conflict, andsome of which are universally arrayed in markedness subhierarchies.

1 Basics of the Analysis

1.1 Grimshaw’s (1995) analysis of English wh-movement and inversion

English has basic subject–verb–object word order. In matrix, nonsubject wh-questions, subject–verb word order is ungrammatical, and inversion (verb–subject word order) is required.7 This isshown by the contrast between (9a) and (9b). The order subject–verb is obligatorily maintained insubordinate wh-questions, as shown in (9c-d).

(9) English matrix and subordinate wh-questions

a. Matrix—inversion What has Michael eaten?b. *Matrix—no inversion * What Michael has eaten?c. *Subordinate—inversion * I wonder what has Michael eaten.d. Subordinate—no inversion I wonder what Michael has eaten.

Within current assumptions about syntactic structure, an analysis of inversion in English wouldhave it that wh-movement, at least that of nonsubjects, triggers inversion in matrix clauses. Morespecifically, nonsubject wh-movement8 creates an environment that prompts movement of the verbover the subject in matrix clauses but not in subordinate clauses.

In Grimshaw’s (1995) analysis, the OP-SCOPE requirement (6a) is not inviolable. Rather,it is built into OP-SPEC, so a wh-phrase taking scope from an adjoined position doesn’t satisfy theconstraint. Grimshaw also assumes an additional violable constraint, OB-HD (for OBligatoryHeaDs), demanding that a projection have a (filled) head. In her system, OP-SPEC and OB-HDtogether trigger inversion, if both dominate STAY. OP-SPEC ranked above STAY forces wh-movement to the specifier position of an otherwise unnecessary functional projection c-commanding the extended projection of the lexical verb; OB-HD ranked above STAY forces thefilling of the head of that functional projection by head movement. The following structure showshow this system works with the English example in (9a).9

7 Though inversion is limited to auxiliary verbs in English. See Grimshaw (1995) for detailed discussion and analysis.8 Since this paper is devoted in its entirety to nonsubject wh-movement, I will henceforth not distinguish it from subjectwh-movement and will call it simply wh-movement.9 The symbol ‘»’ indicates the domination relation in Optimality-theoretic notation. Constraints separated by commaswithin curly brackets are constraints that are not crucially ranked with respect to each other, because they do not directlyconflict and intervening constraints distinguishing their respective ranking are not at issue.


6

(10) {OP-SPEC, OB-HD} » STAY

VP

DP V´

what V VP

has DP V´

Michael tv VP

eaten twh

Constraints satisfied (√) / violated (*)

√OP-SPEC√OB-HD**STAY

The tableau in (11) shows the competition among plausible candidate outputs of the input to (10).

(11) {OP-SPEC, OB-HD} » STAY

Candidates: Matrix clause OP-SPEC OB-HD STAY

a. [VP Michael [V´ has [VP eaten what ] ] ] *!

b. [XP what [X´ ø [VP Michael [V´ has [VP eaten twh ] ] ] ] ] *! *

c. = (10) ☞ [VP what [V´ has [VP Michael [V´ tv [VP eaten twh ] ] ] ] ] * *

Candidate (11a) is the bare verbal projection, with nothing moved. OB-HD is fully satisfied, as isSTAY; but high-ranking OP-SPEC is crucially violated, as indicated by the asterisk. Since the othertwo candidates fare better on OP-SPEC, this candidate loses, as indicated by the exclamation mark.Candidate (11b) satisfies OP-SPEC at the expense of STAY by moving the wh-phrase to thespecifier of some headless XP above VP. This candidate violates high-ranking OB-HD, and soloses. Candidate (11c) satisfies OP-SPEC and OB-HD at the double expense of STAY. This is thegrammatical structure that is diagrammed in (10). Since STAY is crucially low-ranked, theseparticular violations of STAY are irrelevant, as indicated by the shading, to this candidate’sselection as optimal, as indicated by the pointing hand.10

This establishes that OP-SPEC and OB-HD dominate STAY in English. But if this is thecase, how is it that auxiliaries fail to invert in subordinate clauses? Grimshaw’s answer is that theconstraint PROJ-PRIN, given below in (12), dominates OB-HD, forcing violation of OB-HD just insubordinate clauses.11

(12) Constraint regulating subordinate clauses

PROJ-PRIN

No adjunction to subordinate clauses, and no movement into the head of asubordinate clause.

10 Further gratuitous violations of STAY are not tolerated — the optimal candidate here competes against other candidateswith more STAY violations, and wins because it has less. For more discussion of optimal candidate selection in OptimalityTheory, see Prince & Smolensky (1993), especially the Appendix. I purposely abstract away from any possible A-movement of the subject (e.g., from the lower Spec/VP), which may also violate STAY but is independently motivated byGrimshaw’s SUBJ constraint (ranked higher than STAY in English) and so is not relevant here.11 The formulation of PROJ-PRIN is a generalization of proposals by Rizzi & Roberts (1989) and McCloskey (1992).

Eric J. Bakovic

7

The underlined portion of the constraint is what forces the OB-HD violation; to satisfy OB-HD in asubordinate wh-question, there must be movement into the head of the subordinate clause, but thiswould violate PROJ-PRIN, so the movement is blocked if PROJ-PRIN dominates OB-HD. Leavingthe wh-phrase in situ to avoid both the PROJ-PRIN and OB-HD violations is not an option, whichshows that OP-SPEC also dominates OB-HD in English. The tableau in (13) shows this.

(13) {PROJ-PRIN, OP-SPEC} » OB-HD » STAY

Candidates: Subordinate clause PROJ-PRIN

OP-SPEC

OB-HD

STAY

a. [VP Michael [V´ has [VP eaten what ] ] ] *!

b. ☞ [XP what [X´ ø [VP Michael [V´ has [VP eaten twh ] ] ] ] ] * *

c. [VP what [V´ has [VP Michael [V´ tv [VP eaten twh ] ] ] ] ] *! * *

The following subsection argues for an alternative to this analysis, in which OB-HD is not aviolable constraint in the theory but rather an inviolable principle of X´ Theory. This proposalforms the core of the analysis of inversion in Spanish.

1.2 wh-movement and inversion without OB-HD

Consider the structures in (14), where YP and X have both been moved.

(14) X´ Structures

a. XP

YP X´

ZP

b. XP

YP X´

X ZP

c. ZP

YP ZP

*OB-HD, *STAY **STAY *OP-SPEC, *STAY

Suppose YP in (14) is a syntactic operator; a wh-phrase. In (14a), YP has satisfied OP-SPEC bymoving from ZP to the specifier of a (literally) headless projection XP above ZP, violating OB-HDand STAY once each. In (14b), YP has satisfied OP-SPEC by moving from ZP to the specifier of aprojection XP headed by some X above ZP, satisfying OB-HD by inverting some head into X,violating STAY once for each of the two movements. In (14c), YP has violated OP-SPEC andSTAY once each by moving from ZP to a position adjoined to ZP, vacuously satisfying OB-HD bycreating no projection in need of a head.

Grimshaw (1995:§8), assuming that adjunction of wh-phrases as in (14c) is not an option,argues that (14a) is the representation of a wh-question with no inversion (such as a subordinatewh-question in English) and that (14b) is the representation of a wh-question with inversion (suchas a matrix wh-question in English). I argue instead that there is no such thing as a headlessprojection like (14a). I propose that instead of OB-HD, the requirement that a projection have ahead is an inviolable principle of X´ Theory. Maximal projection movements without anobligatorily following head, then, are all cases of adjunction to a maximal projection, as in (14c).

This type of structure satisfies OP-SCOPE but violates OP-SPEC, which I assume to beseparate requirements (the latter being a violable constraint and the former being an inviolableprinciple of interpretation; see §1.4.2). Since matrix wh-questions in English display inversion,


8

wh-movement in this case is movement to specifier as in (14b), satisfying OP-SPEC. Sincesubordinate wh-questions in English do not display inversion, wh-movement in this case isadjunction as in (14c), in violation of OP-SPEC. The broad claim made here is that all maximalprojection movements that require a following head are movements to specifer, while all maximalprojection movements that do not require a following head are adjunctions.

Under this view, both wh-movement and inversion are explained simply by ranking OP-SPEC above STAY. For the operator to be in a c-commanding specifier position to satisfy OP-SCOPE and OP-SPEC, it must move to a projection with a head. The movement of the operatorviolates STAY. For there to be a head in this projection, a head must move into it. This movementalso violates STAY. Satisfaction of OP-SPEC can thus force two violations of STAY.

The tableau in (16) shows the competition between two plausible candidate outputs of theinput to (10). The difference between tableau (16) and tableau (11) is the absence in (16) of (a)OB-HD (OB-HD is not a violable constraint), (b) a “headless projection” candidate violating OB-HD(because that candidate would not be a legitimate X´ structure), and (c) a candidate with no wh-movement (because that violates inviolable OP-SCOPE). Instead, a candidate is considered withadjunction of the wh-phrase.

(16) OP-SPEC » STAY

Candidates: Matrix clause OP-SPEC STAY

inv ☞ [VP what [V´ has [VP Michael [V´ tv [VP eaten twh ] ] ] ] ] * *

adj [VP what [VP Michael [V´ has [VP eaten twh ] ] ] ] *! *

The first candidate in (16), labelled inv, shows wh-movement to specifier and inversion, satisfyingOP-SPEC at the double expense of STAY. The second candidate, labelled adj, shows wh-adjunction and no inversion, satisfying OP-SCOPE, the inviolable principle of interpretation, at theexpense of STAY, and avoiding a second violation of STAY at the expense of OP-SPEC. As thepointing hand shows, inv wins, because it best satisfies the constraint hierarchy: only low-rankedSTAY is violated by it.

The absence of inversion in English subordinate clauses is understood here as being due toforced adjunction of wh-phrases to subordinate clauses. What could force this violation of PROJ-PRIN (12), half of which demands no adjunction to subordinate clauses? Grimshaw notes that “itis possible that PROJ-PRIN is properly understood as two constraints, one on adjunction and oneon head movement, which both regulate subordinate clauses. In this case, it will be possible torank them separately” (p. 23).

It seems that this separation is necessary in this OB-HDless system (in some sense; see§1.4.3). To avoid moving into the head of a subordinate clause and violating one half of PROJ-PRIN, it is necessary to adjoin the wh-phrase to the subordinate clause, violating the other half.This is because movement into the head of a subordinate clause, which would be forced by wh-movement to specifier to satisfy OP-SPEC, would violate what I call PROJ-PRIN(H) — the half ofPROJ-PRIN demanding no movement into the head of a subordinate clause. This establishes that inEnglish, PROJ-PRIN(H) dominates OP-SPEC and the other half of PROJ-PRIN, which I call PROJ-PRIN(A), since it forces violation of each. This is shown by the tableau in (17), showing thesubordinate clause versions of the same candidates as (16).12

12 I occasionally use P-P(H) and P-P(A) as abbreviations for PROJ-PRIN(H) and PROJ-PRIN(A), respectively.

Eric J. Bakovic

9

(17) PROJ-PRIN(H) » {OP-SPEC, PROJ-PRIN(A)}

Candidates: Subordinate clause P-P(H) OP-SPEC P-P(A)

inv [VP what [V´ has [VP Michael [V´ tv [VP eaten twh ] ] ] ] ] *!

adj ☞ [VP what [VP Michael [V´ has [VP eaten twh ] ] ] ] * *

This time, adj wins because it best satisfies the relevant constraint hierarchy. Highest-rankedPROJ-PRIN(H) forces violation of OP-SPEC and PROJ-PRIN(A) with wh-adjunction, since wh-movement to specifier and inversion will violate PROJ-PRIN(H) (as shown by the failure of inv).13

The immediate benefit of dispensing with OB-HD as a violable constraint is that it cannot bedifferentially ranked with other constraints, enlarging the set of possible rankings and thus the setof possible grammars. For instance, Grimshaw (1995:§7) notes that the ranking in (18) predicts agrammar with wh-movement and inversion in matrix clauses and neither in subordinate clauses.This does not, as far as either Grimshaw or I are aware, correspond to any known language.

(18) Problematic ranking

PROJ-PRIN » OB-HD » OP-SPEC » STAY

Because PROJ-PRIN blocks movement into the head of a subordinate clause in this system by beingranked above OB-HD (as in English), OB-HD will be violated whenever there is subordinate clausewh-movement compelled by OP-SPEC. Ranking OB-HD above OP-SPEC avoids this violation ofOB-HD, by blocking wh-movement in subordinate clauses. This is shown by the contrast in (19)between matrix and subordinate clause wh-movement under this ranking, using English candidatesfor easy comparison with the tableaux seen so far.

(19) PROJ-PRIN » OB-HD » OP-SPEC » STAY

a. Candidates: Matrix clause PROJ-PRIN

OB-HD

OP-SPEC

STAY

[VP Michael [V´ has [VP eaten what ] ] ] *!

[XP what [X´ ø [VP Michael [V´ has [VP eaten twh ] ] ] ] ] *! *

☞ [VP what [V´ has [VP Michael [V´ tv [VP eaten twh ] ] ] ] ] * *

b. Candidates: Subordinate clause PROJ-PRIN

OB-HD

OP-SPEC

STAY

☞ [VP Michael [V´ has [VP eaten what ] ] ] *

[XP what [X´ ø [VP Michael [V´ has [VP eaten twh ] ] ] ] ] *! *

[VP what [V´ has [VP Michael [V´ tv [VP eaten twh ] ] ] ] ] *! * *

At the center of this problematic ranking is OB-HD, struggling to be satisfied under compulsion ofPROJ-PRIN. With the inviolable OP-SCOPE principle forcing (at least) wh-adjunction, and with thesplitting of PROJ-PRIN into separate constraints, the need for OB-HD is obviated and this problemall but vanishes. (See §§1.4.2–1.4.3 for more on OP-SCOPE and the splitting of PROJ-PRIN.)

13 Since it has already been established that STAY is ranked below OP-SPEC, it is irrelevant to the decision between thecandidates in this case and so has been omitted for visual convenience.


10

I have shown here that the same results that Grimshaw obtains with OB-HD as a violableconstraint can be obtained with it as an inviolable principle of X´ Theory. Since satisfaction of OP-SPEC (or any constraint that requires movement to specifier and hence inversion) entailssatisfaction of OB-HD in this view, then OP-SPEC itself can do the job in constraint rankings thatOB-HD does in Grimshaw’s system. OB-HD was only active in Grimshaw’s system if forced to beby another constraint, specifically OP-SPEC. Thus, nothing more needs to be said about how toreanalyze/mimic the work done by OB-HD when it is embedded in a constraint hierarchy.

How a language chooses to satisfy the now inviolable requirement of OB-HD in a givencontext will be entirely determined by the ranking of other constraints. For instance, if it issatisfied by inversion, then STAY is ranked lower than other constraints that can be violated byhead-fillers; if it is satisfied by direct insertion of a dummy head, then FULL-INT, the constraintagainst meaningless elements, is the lowest-ranked of such constraints; etc. Complex interactionsof these and other constraints can yield “mixed” systems like English with auxiliary inversion anddo support with main verbs in matrix but not subordinate wh-questions, as Grimshaw (1995) hasalready successfully shown.

The broad claim underlying this proposal, that all movements of maximal projections thatrequire a following head are movements to specifier and that all movements of maximal projectionsthat do not require a following head are adjunctions, has obvious consequences for any theory thatincorporates Spec-Head “feature checking” constraints; constraints of the general form “be in somespecifier” like OP-SPEC. If movement motivated by what seems to be (s-structure) featurechecking does not require a following head, then it is not movement to specifer, it is adjunction.Thus, this movement cannot be motivated by feature checking in any strict sense (there is no directSpec-Head relation), and its motivation must be found elsewhere.

Any maximal projection movements that are substitutions in some contexts and adjunctionsin others, like the matrix/subordinate wh-question distinction in English, will need to be motivatedby a “be in some specifier” constraint and an OP-SCOPE-type principle (or similar constraint,ranked highly enough). The soft prediction made by the present proposal, then, is that suchmovements will only be movements of syntactic operators or other elements that require syntacticscope. I leave this prediction open for future evaluation.14

1.3 Spanish inversion and different operator types

The previous subsection explained how the inversion pattern in English wh-questions can beexplained without using Grimshaw’s OB-HD constraint. Now we turn to Spanish. Spanish hasbasic subject–verb–complement/adjunct word order, as shown in (20).

(20) Basic subject–verb–complement/adjunct word order

a. S–V–Complement Miguel [se comió]V la manzana.M ate.3s the apple

‘Miguel ate the apple.’b. S–V–Adjunct Miguel [se fue]V a casa.

M went.3s to house‘Miguel went home.’

14 Topicalization may be a fruitful avenue for research in this area, as suggested to me by Bruce Tesar and Paul Smolensky.

Eric J. Bakovic

11

Inversion is, depending on the dialect, the clause-type and the argumental nature of the wh-phrase,obligatory in wh-questions.15 Consider the dialect in (1f), which I call Spanishƒ for convenience.Spanishƒ requires inversion with arguments but not adjuncts in matrix clauses, and with neither insubordinate clauses, as shown by the examples in (21) and (22).16

(21) Spanishƒ matrix wh-questions

a. Arg–V–S ¿Qué [se comió]V Miguel?what ate.3s M‘What did Miguel eat?’

b. *Arg–S–V * ¿Qué Miguel se comió?

c. Adj–S–V ¿Dónde Miguel [se fue]V?where M went.3s

‘Where did Miguel go?’

(22) Spanishƒ subordinate wh-questions

a. Arg–S–V Me pregunto qué Miguel [se comió]V.wonder.1s what Miguel ate.3s

‘I wonder what Miguel ate.’b. Adj–S–V Me pregunto dónde Miguel [se fue]V.

wonder.1s where Miguel went.3s‘I wonder where Miguel went.’

Suppose that the only distinction between operator types is the one made in Spanishƒ, betweenargument wh-phrases (Arg) and adjunct wh-phrases (Adj). From the data in (21), we know that inmatrix clauses, Arg triggers inversion while Adj does not. In terms of the basic analysis of §1.2above, Arg satisfies OP-SPEC at the expense of an extra violation of STAY, the head movement thatis not independently motivated by OP-SCOPE, while Adj violates OP-SPEC to better satisfy STAY.

The conclusion is that each operator type has a separate OP-SPEC constraint referring to it:ARGOP-SPEC and ADJOP-SPEC, respectively. ARGOP-SPEC must dominate STAY, since Argmust move to a specifier position and trigger inversion, forcing two STAY violations. On the otherhand, STAY must dominate ADJOP-SPEC, since Adj must adjoin and not trigger inversion, forcingonly one STAY violation (due to the inviolable OP-SCOPE).

The tableaux in (23) below show how decisions of optimality are made in Spanishƒ matrixwh-questions. The two relevant candidates in each case are the now familiar inv and adj. Tableau(23a) shows what happens with Arg. Since ARGOP-SPEC dominates STAY, inv wins: it is betterto violate STAY a second time than to not have the operator in specifier position (ADJOP-SPEC isvacuously satisfied; there is no Adj in this case). Tableau (23b) shows what happens with Adj.Since STAY dominates ADJOP-SPEC, adj wins: it is better to not have the operator in specifierposition than to violate STAY a second time (ARGOP-SPEC is vacuously satisfied).

15 Verb–subject word order is in fact always possible, even in declaratives; it is known as free inversion (see fn. 1). Forthis reason, verb–subject word order is not considered when the subject–verb word order that results from wh-adjunction isgrammatical. See the analysis in §4 of “free” inversion.16 An inverted question mark ‘¿’ before an example is in accord with Spanish orthography and not a grammaticalityjudgment. I leave some space between a judgment symbol and the sentence it corresponds to, as in (21b).


12

(23) ARGOP-SPEC » STAY » ADJOP-SPEC

a. Candidates: Matrix clause, Argument operator ARGOP-SPEC

STAYADJOP-SPEC

inv ≈ (21a) ☞ [VP Arg [V´ V [VP Subj [V´ tv twh ] ] ] ] * *

adj ≈ (21b) [VP Arg [VP Subj [V´ V twh ] ] ] *! *

b. Candidates: Matrix clause, Adjunct operator ARGOP-SPEC

STAYADJOP-SPEC

inv [VP Adj [V´ V [VP Subj [V´ tv twh ] ] ] ] **!

adj ≈ (21c) ☞ [VP Adj [VP Subj [V´ V twh ] ] ] * *

With PROJ-PRIN(H) ranked above ARGOP-SPEC, the Spanishƒ subordinate wh-question examplesin (22) are predicted. Since PROJ-PRIN(H) dominates ARGOP-SPEC by hypothesis and ADJOP-SPEC by transitivity, adj wins in subordinate clauses for both Arg (24a) and Adj (24b). It is betterto not have either operator in specifier position than to violate PROJ-PRIN(H).17

(24) PROJ-PRIN(H) » ARGOP-SPEC » STAY » ADJOP-SPEC

a. Candidates: Subordinate clause, Argumentoperator P-P(H) ARGOP-

SPECSTAY

ADJOP-SPEC

inv [VP Arg [V´ V [VP Subj [V´ tv twh ] ] ] ] *! * *

adj ≈ (22a) ☞ [VP Arg [VP Subj [V´ V twh ] ] ] * *

b. Candidates: Subordinate clause, Adjunctoperator P-P(H) ARGOP-

SPECSTAY

ADJOP-SPEC

inv [VP Adj [V´ V [VP Subj [V´ tv twh ] ] ] ] *! * *

adj ≈ (22b) ☞ [VP Adj [VP Subj [V´ V twh ] ] ] * *

Because other Spanish dialects make the distinction between operator types that trigger inversionand those that don’t in places other than the argument/adjunct juncture, OP-SPEC must be separatedinto different constraints that make the OP-SPEC demand on each of the different operator types.These constraints are universally arrayed in a markedness subhierarchy, reflecting the most-to-leastargumental markedness scale introduced in §0 and discussed in more detail in §2.

In that section I also show how the differences between Spanish dialects with respect toinversion are generally explained by ranking STAY in different spots of the markednesssubhierarchy. Those members of the markedness subhierarchy ranked above STAY force twoviolations of STAY as in (23a), while those ranked below STAY cannot force any violations ofSTAY. Only inviolable OP-SCOPE can, and it does so minimally through adjunction of theoperator, as in (23b). In §3, I take up the more specific issue of the matrix/subordinate distinctionas previewed in (24) vs. (23). But first, allow me to clarify some basic assumptions.

17 This simple case shows trivial allegiance to the subset relationship (3); the more complete proof that this subsetrelationship is obeyed under any possible ranking of the constraints proposed is given in §3.

Eric J. Bakovic

13

1.4 Some assumptions18

1.4.1 Input and output

Now is a good time to discuss in some detail the inputs to the competitor sets (pairs, actually)being evaluated by the constraint hierarchies in the tableaux in this paper. Following Grimshaw(1995), Samek-Lodovici (in preparation) and Grimshaw & Samek-Lodovici (1995), I assume theinput of a clause to consist minimally of a triple (L, M, A) such that:

(i) L is a lexical head with its argument structure.(ii) M is an optional marking of the arguments of L as foci, or topic-referring (see §4).(iii) A is an assignment of lexical heads to L’s arguments.

The input-output function Gen then generates from the input triple all possible L-headed syntacticstructures S such that:

(iv) S obeys X´ Theory.(v) Segments and projections are generated freely; projections [and heads] may move

[i.e., may be overtly realized anywhere], leaving a [coindexed] trace behind [i.e., intheir “base-generated” position, to satisfy the inviolable Theta Criterion].

From Samek-Lodovici (in preparation).

I also assume that scope-taking elements such as wh-phrases, quantifiers and the like are specifiedin the input for where their scope is taken from. For example, the input to a simple clause with awh-phrase complement is as in (25a), and the input to a clause with a clausal complement whichitself has a wh-phrase complement (taking narrow scope) is as in (25b). The narrow scope of thewh-phrase in (25b) is represented by coindexation with the lexical head of the lower clause.19

(25) Inputs

a. (V<x,y>, <x=j>, <y=wh>)b. (V<x,y>, <x=j>, <y=(Vi<x,y>, <x=m>, <y=whi>)>)

The output of every input is determined by the constraint hierarchy of the language in question.Using English, if the lexical head V in (25a) is eat, j is John, and wh is what, then this is the inputto the direct question What did John eat? (a matrix wh-question). If the first lexical head V in(25b) is wonder, the second is eat, j is John, m is Mary, and wh is what, then this is the input tothe indirect question John wonders what Mary ate (a subordinate wh-question). These are the twoinput-types considered in this paper (with focus entering the picture in §4).

The requirement that “S obeys X´ Theory” in (iv) above of course includes my proposal in§1.2 to dispense with a violable OB-HD, making compliance with that constraint an inviolableprinciple of X´ Theory. In addition, following the claim that OP-SCOPE is an inviolable principleof interpretation, I assume that wh-elements always obey OP-SCOPE by moving to an appropriatec-commanding position, syntactically reflecting their interpretive scope. I turn to this point now.

18 For encouraging me to clear up the issues addressed in this subsection, I’d like to thank Viviane Déprez (§1.4.1),Veneeta Dayal (§1.4.2) and Jane Grimshaw (§1.4.2 and §1.4.3).19 Arguments are not marked here as foci or as topic-referring for simplicity.


14

1.4.2 OP-SCOPE

If OP-SCOPE is an inviolable principle of interpretation, it will always force wh-movement orcoindexation of in situ (or partially moved) wh-phrases with clause-initial operators. If this iscorrect, then wh-in-situ languages will have to be analyzed as obeying OP-SCOPE along the linesof Aoun & Li’s (1993ab) proposal that a sometimes overt, sometimes covert operator, coindexedwith the in situ wh-phrase, is present clause-initially in Chinese and Japanese (see also Cheng1991 and Li 1992). Other wh-in-situ phenomena, such as that observed in multiple wh-questionsin languages like English, must be analyzed as obeying OP-SCOPE through coindexation of the insitu wh-phrase(s) with the moved one (absorption; see Higginbotham & May 1981). Partial wh-movement in languages like German, as analyzed in McDaniel (1989), involves coindexation of thepartially moved wh-phrase with an overt scope marker, satisfying OP-SCOPE.

See Legendre et al. (1995) for a proposal within OT (but using somewhat differentconstraints) accounting for the differences among wh-in-situ languages (their example is Chinese),single wh-movement languages (English), and multiple wh-movement languages (Bulgarian). Seealso the work of Ackema & Neeleman (1995). The issues involved in the characterization of thesedifferent languages using the present constraints are unfortunately beyond the scope of this paper.

1.4.3 PROJ-PRIN(A)20

If PROJ-PRIN(A) is allowed to be ranked above PROJ-PRIN(H) and STAY, and if STAY is in turnranked above OP-SPEC,21 then the subset relationship (3) can be violated: subordinate clauses canrequire inversion with more wh-phrase types than matrix clauses. The tableaux in (26) show why,with a relevant total ranking: tableau (26a) shows the case for matrix clauses, where the PROJ-PRIN constraints are irrelevant; (26b) for subordinate clauses, where the PROJ-PRIN constraintsmake the decisions.

(26) PROJ-PRIN(A) » PROJ-PRIN(H) » STAY » OP-SPEC

a. Candidates: Matrix clause P-P(A) P-P(H) STAY OP-SPEC

inv [VP wh [V´ V [VP Subj [V´ tv twh ] ] ] ] **!

adj ☞ [VP wh [VP Subj [V´ V twh ] ] ] * *

b. Candidates: Subordinate clause P-P(A) P-P(H) STAY OP-SPEC

inv ☞ [VP wh [V´ V [VP Subj [V´ tv twh ] ] ] ] * * *

adj [VP wh [VP Subj [V´ V twh ] ] ] *! * *

As shown in §3, the interactions among only PROJ-PRIN(H), STAY and OP-SPEC are bothnecessary to guarantee the subset relationship (3) and sufficient to decide between the two relevantcandidates, inv and adj, in matrix and subordinate clauses. This means that the ranking in (26)must somehow be avoided.

I suggest here three alternatives for avoiding this problem. One alternative is that PROJ-PRIN(A) is simply low-ranked (below PROJ-PRIN(H), at least) in the grammars of all Spanishdialects. This is not interesting because it doesn’t explain the universality of the subset relationship

20 I thank Jane Grimshaw and Paul Smolensky for very helpful discussions concerning the entire issue of PROJ-PRIN(A).21 Since there is more than one OP-SPEC constraint, this should say more precisely “… ranked above the relevant memberof the OP-SPEC family”. I take this obvious point for granted in what follows and at various other points of this paper.

Eric J. Bakovic

15

(3). Another alternative is that PROJ-PRIN(A) is universally lower-ranked than PROJ-PRIN(H).Constraints that are universally ranked with respect to each other are of course not unheard of; I amproposing a set of those myself in this paper in the form of a markedness subhierarchy. However,the constraints in the markedness subhierchy make the same demand (the OP-SPEC demand) ondifferent linguistic objects (different types of wh-phrases), while the PROJ-PRIN constraints makedifferent demands (no movement into the head, no adjunction) on the same linguistic object(subordinate clauses). There is no foolproof diagnostic for whether a pair of constraints arecandidates for universal ranking with respect to each other, but this difference is striking.

Both of these alternatives try to avoid the problem by keeping PROJ-PRIN(A) as a constraintseparate from PROJ-PRIN(H), but subverting its activity somehow. As Paul Smolensky haspointed out to me in a personal communication, one would like to find evidence that furthersupports the separation of the two constraints in a given grammar, as Grimshaw (1995:23)originally noted. This evidence would have to consist of a third constraint that crucially intervenesbetween them, able to force violation of the lower one (presumably PROJ-PRIN(A)) and in turn ableto be violated under pressure by the higher one (PROJ-PRIN(H)). I am not aware of any suchevidence at present.

The third, most interesting alternative is that PROJ-PRIN(A) and PROJ-PRIN(H) really areone constraint, as Grimshaw originally proposed, but that violation of the PROJ-PRIN(H) part ofthe constraint is somehow worse than violation of the PROJ-PRIN(A) part of the constraint, at leastwhen it comes to movement of the sort being investigated here. This alternative could build on thefact that wh-movement and inversion, a violation of PROJ-PRIN(H), creates two levels of structureon top of a subordinate clause (one X´-level, one XP-level) while adjunction, a violation of PROJ-PRIN(A), creates only one (an XP-level). This interpretation of the once again unified PROJ-PRINconstraint would thus be parallel to STAY (one violation for adjunction, two for inversion), andcould perhaps be completely reduced to a more specific version of STAY in this sense. However,if STAY governs A-movement as well as A´- and head-movement in OT, as assumed by Grimshaw(1995), then this reduction is probably too strong, as it predicts a vast number of (unattested, itseems) matrix/subordinate distinctions with respect to movement.

All of this is a reflection of the poor understanding we have of the specific formulation ofthe PROJ-PRIN constraint(s), and of the exact nature and extent of matrix/subordinate distinctions.However, the claim that there is/are constraint(s) regulating just subordinate clauses and not justmatrix clauses, and that violating one (part) of the constraint(s) is worse than violating the other,satisfies a necessary condition of the present analysis: that the matrix/subordinate distinction, inparticular the subset relationship (3), be captured. Satisfaction of sufficient conditions demandingcomplete explanatory adequacy in the form of perfectly formulated constraints is, I think, importantbut secondary to this goal. I leave this issue for future research to shed light on.

2 The Markedness Scale of Operator Types

Spanish dialects vary as to what type of wh-phrase triggers inversion. In this paper I restrictmyself to the simple operators quién ‘who’, qué ‘what’, dónde ‘where’, cuándo ‘when’, cómo‘how’, and por qué ‘why’. Future research will connect the analysis of the facts involving theseelements with the facts involving other operators.

I have conducted a cross-dialectal judgment survey and have found that in all dialects thatexhibit inversion in matrix wh-questions, it is required with the argument operators quién ‘who’and qué ‘what’. A subset of these dialects also requires inversion with the (spatio-temporal)location operators dónde ‘where’ and cuándo ‘when’. A subset of this subset also demandsinversion with the manner operator cómo ‘how’. Finally, a subset of this latter subset requires


16

inversion with all operators, the only other one considered here being the reason operator por qué‘why’. With respect to inversion in wh-questions, then, the picture of Spanish dialects thatemerges is (27), repeated from (2).

(27) Implicational relationship among wh-phrase types

A matrix/subordinate wh-question requires inversion if the wh-phrase is

1(Torrego 1984)

dónde, cuándo cómo por quéquién, qué‘how’ ‘why’

4(Survey results)

2(Survey results)

3(Goodall 1991ab)

‘who, what’ ‘where, when’

ARGUMENT LOCATION MANNER REASON

0 = no inversion (Suñer 1994)

Pattern 1, with inversion only required with who and what, corresponds to the dialect analyzed inTorrego (1984) and Suñer (1994).22 Pattern 3, with inversion required with everything but why,is the dialect analyzed in Goodall (1991ab). Patterns 2 and 4 have emerged from my survey. Afifth pattern, which I refer to as pattern 0, requires no inversion at all. This corresponds to anotherdialect discussed in Suñer (1994).

This implicational relationship runs from what I am calling most to least “argumental”, witharguments being on the more argumental side and reasons being on the less argumental side. Whyis the space of possibilities divided up in just this way? This question is of course very difficult toanswer, and I will not even attempt to answer it here in any fundamental way. However, there areobvious precedents in the literature to the markedness scale uncovered here, and I make referenceto those precedents here.

Distinctions very similar to the ones made among the operator types above (argument,location, manner, and reason) are made in the recent literature under the general heading ofreferentiality (see in particular Rizzi 1990 and Cinque 1990), following Huang’s (1982) seminalanalysis of the differences between arguments and adjuncts with respect to long-distanceextractability. Legendre et al. (1995) interpret the distinctions made in the literature “on a hierarchyof participants to an event with central participants (agent, patient [≈ arguments — E.J.B.]) at oneend and exterior conditions at the other (manner, reason). The grammar of a particular languageselects a cut-off point on the hierarchy which dichotomizes it into two parts which the syntax treatsdifferently” (p. 21).

Once seen in this scalar way, rather than as an arbitrary-sounding distinction between‘referential’ and ‘nonreferential’, this referentiality hierarchy seems to be the same one underlyingthe argumental markedness scale uncovered here. In particular, arguments are on one end whilemanner and reason are on the other. This indicates the existence of a general referentialityhierarchy to which the syntax of each language is differentially sensitive.

But are they indeed the same? The cut-off points in the markedness scale that the differentSpanish dialects make affect the syntax of inversion, rather than the syntax of long-distanceextraction that is sensitive to the referentiality hierarchy. However, it is not predicted that eachlanguage will make one and only one cut-off point in any hierarchy or markedness scale, as may beimplied here. For instance, there are crucial differences, in terms of the exact constraintsresponsible, between the analysis of inversion detailed below and the analysis of long-distance

22 I occasionally represent Spanish forms with their English glosses in italics to simplify the text.

Eric J. Bakovic

17

extraction given in Legendre et al. (1995). Like the sonority scale in phonology, the referentialityhierarchy may simply underlie various independent markedness subhierarchies.23,24

Though the parallels here are striking and worthy of further investigation, I hesitate to claimthat one and the same referentiality hierarchy underlies the syntax of both long-distance extractionand inversion. Of course, I do not rule out this possibility — I only wish to see more than mymeager sample of simple operators put to the test of a rigorous judgment survey, taking intoaccount all aspects known or suspected to be related to referentiality: argument structure, theta-roles, discourse-linking, connectedness to the predicate, and so on. Until this is done, I must relyonly on my limited findings, solid though they may be, and refer to the scale as more-or-less“argumental”, to emphasize my ambivalence about jumping on the referentiality bandwagon.

As noted by Legendre et al. (1995): “A proper treatment of the referentiality hierarchy, [orany hierarchy — E.J.B.] and typological variation in the cut-off point, should be naturally handledby OT (much as the sonority hierarchy in Prince and Smolensky 1993)” (p. 22). This reference isto the significant advance in the understanding of markedness in phonology that was made byPrince & Smolensky (1993) within OT. Their general proposal was that the individual members ofcertain constraint families (sets of constraints making the same demands on different types oflinguistic objects) are universally ranked with respect to each other, defining a constraintsubhierarchy that reflects a markedness scale, or a markedness subhierarchy. Other constraints canbe parochially ranked with respect to the individual members of the markedness subhierarchy,predicting all and only language types conforming to the markedness scale.

The fact that no dialect of Spanish divides the world of operators any differently than in(27) by itself implies the existence of such a markedness scale; the fact that it parallels similardistinctions in other languages merely enforces the implication. It can be said that in Spanish,argument operators require inversion more than location operators, which require it more thanmanner operators, which require it more than reason operators. In terms of the analysis of wh-movement and inversion in §1, an argument wants to be in specifier position more than a locationmore than a manner more than a reason, since movement to specifier is what triggers inversion.

To account for the Spanish facts with respect to the observed markedness scale, I proposethat each type of operator (argument, location, etc.) has its own OP-SPEC constraint, definedindividually in (28a–d), repeated from (7a-d). These constraints are members of the OP-SPECconstraint family, which defines a markedness subhierarchy because its members are universallyranked with respect to each other as in (28e), repeated from (7e).25 I will henceforth refer to (28e)as “the markedness subhierarchy.”

23 I thank Paul Smolensky for pointing this out to me.24 In the context of argument/adjunct asymmetries with respect to long-distance extractability, Rizzi (1990) argues that areferential theta-role is indexed and can thus satisfy the ECP through binding, while nonreferential theta-roles are notindexed and thus must satisfy the ECP through the more local government relation. In this theory, UG distinguishesindexed from non-indexed theta-roles, but the grammar of each language must distinguish referential from nonreferentialtheta-roles because this is variable. By separating referentiality from indexing in this way, Rizzi inadvertently predictsthat each language will make one and only one cut-off point in the referentiality hierarchy.25 Paul Smolensky points out to me in a personal communication that the markedness subhierarchy responsible fordifferences in long-distance extractability can be understood as the local conjunction (see Smolensky 1995 for a definition)of a more general referentiality hierarchy and the MINLINK family of constraints. Though not decomposed as such here,(28e) may ultimately be understood as the local conjunction of the referentiality hierarchy and OP-SPEC. In this way, thereferentiality hierarchy parallels the sonority hierarchy in phonology, since the latter hierarchy is the basis of a number ofconstraint families in phonology (through local conjunction, according to Smolensky), many if not all of which areassumed to be arrayed in markedness subhierarchies following Prince & Smolensky (1993:§9).


18

(28) OP-SPEC constraint family

a. ARGOP-SPEC

Argument operators must be in specifierposition.

b. LOCOP-SPEC

Location operators must be in specifierposition.

c. MANOP-SPEC

Manner operators must be in specifierposition.

d. REASOP-SPEC

Reason operators must be in specifierposition.

e. Markedness subhierarchyARGOP-SPEC » LOCOP-SPEC » MANOP-SPEC » REASOP-SPEC

A conflicting constraint like STAY (and PROJ-PRIN(H); see §3) can be ranked anywhere in thissubhierarchy. As demonstrated with Spanishƒ in §1.3, any of the constraints in (28) that is rankedabove STAY will force two violations of STAY to be satisfied. All operators of the type whose OP-SPEC constraint is ranked below STAY will be moved, because of inviolable OP-SCOPE, but to anadjoined position rather than to a specifer, so as to only violate STAY once. Below we examine theranking needed for each of the dialects in turn.

2.2 Typological effects with the markedness subhierarchy 1: STAY

The five different possible ways to rank STAY in the markedness subhierarchy are given below in(29). Each ranking predicts the grammar of the pattern of inversion described below it, also usingthe now familiar numbering convention. The parochial ranking of STAY in each case is highlightedwith shadowed characters.

(29) Typological effects with the markedness subhierarchy

a. SSTTAAYY » ARGOP-SPEC » LOCOP-SPEC » MANOP-SPEC » REASOP-SPEC

⇒ Adjunction of all operators (pattern 0)

b. ARGOP-SPEC » SSTTAAYY » LOCOP-SPEC » MANOP-SPEC » REASOP-SPEC

⇒ Inversion only with argument operators (pattern 1)

c. ARGOP-SPEC » LOCOP-SPEC » SSTTAAYY » MANOP-SPEC » REASOP-SPEC

⇒ Inversion only with argument and location operators (pattern 2)

d. ARGOP-SPEC » LOCOP-SPEC » MANOP-SPEC » SSTTAAYY » REASOP-SPEC

⇒ Inversion with all but reason operators (pattern 3)

e. ARGOP-SPEC » LOCOP-SPEC » MANOP-SPEC » REASOP-SPEC » SSTT AA YY

⇒ Inversion with all operators (pattern 4)

The following subsections explain in detail how the predictions are borne out for each case in (29).Since we are not considering PROJ-PRIN(H) until §3, the predictions are made for the five dialectswith no matrix/subordinate distinction. The remaining dialects are left to be explained along withthe matrix/subordinate distinction in §3.

2.2.1 Pattern 0: no inversion

As shown by the data in (D0), all wh-movements without inversion in matrix and subordinate wh-questions in the dialect with pattern 0 are grammatical.

Eric J. Bakovic

19

(D0) Pattern 0: Data

a. Arg–S–V (Me pregunto) qué Miguel [se comió]V(wonder.1s) what M ate.3s

‘What did Miguel eat? / I wonder what Miguel ate.’b. Loc–S–V (Me pregunto) dónde Miguel [se fue]V

(wonder.1s) where M went.3s‘Where did Miguel go? / I wonder where Miguel went.’

c. Man–S–V (Me pregunto) cómo Miguel [se comportó]V(wonder.1s) how M behaved.3s

‘How did Miguel behave? / I wonder how Miguel behaved.’d. Reas–S–V (Me pregunto) por qué Miguel [se enojó]V

(wonder.1s) why M got-mad.3s‘Why did Miguel get mad? / I wonder why Miguel got mad.’

Ranking STAY above all of the OP-SPEC constraints as in (29a) explains the facts in this dialect.No operator can violate STAY twice just to be in specifier position. However, OP-SCOPE demandsthat the operator move. STAY is a violable constraint, but violation of a constraint undercompulsion of higher-ranked constraints is necessarily minimal in OT: the less violations of alower-ranked constraint it takes to satisfy a higher-ranked constraint, the better (see fn. 10). Tosatisfy OP-SCOPE, all it takes is adjunction of the operator, violating STAY the least one can; once.Any attempt to satisfy one of the OP-SPEC constraints by putting the operator in a specifier positionand requiring head movement violates STAY one too many times.

The constraint tableaux in (T0) show how decisions of optimality are made in this dialect,using the data in (D0). Because STAY outranks all of the OP-SPEC constraints, adj always wins.26

(T0) Pattern 0: Tableaux

a. Candidates: Argument operator STAYARGOP-

SPEC

LOCOP-

SPEC

MANOP-

SPEC

REASOP-

SPEC

inv [VP qué [V´ se comió [VP Miguel [V´ tv twh ] ] ] ] **!

adj ☞ [VP qué [VP Miguel [V´ se comió twh ] ] ] * *

b. Candidates: Location operator STAYARGOP-

SPEC

LOCOP-

SPEC

MANOP-

SPEC

REASOP-

SPEC

inv [VP dónde [V´ se fue [VP Miguel [V´ tv twh ] ] ] ] **!

adj ☞ [VP dónde [VP Miguel [V´ se fue twh ] ] ] * *

26 I adopt the following practice with these demonstrative tableaux: constraint columns corresponding to OP-SPEC

constraints that are irrelevant to the candidates at hand (because the operator type they refer to is not at issue) are shaded —these constraints are vacuously satisfied. The other (relevant) constraint columns are boxed in thicker lines.


20

c. Candidates: Manner operator STAYARGOP-

SPEC

LOCOP-

SPEC

MANOP-

SPEC

REASOP-

SPEC

inv [VP cómo [V´ se comportó [VP Miguel [V´ tv twh ] ] ] ] **!

adj ☞ [VP cómo [VP Miguel [V´ se comportó twh ] ] ] * *

d. Candidates: Reason operator STAYARGOP-

SPEC

LOCOP-

SPEC

MANOP-

SPEC

REASOP-

SPEC

inv [VP por qué [V´ se enojó [VP Miguel [V´ tv twh ] ] ] ] **!

adj ☞ [VP por qué [VP Miguel [V´ se enojó twh ] ] ] * *

2.2.2 Pattern 1: inversion with argument operators

As shown by the data in (D1), inversion is obligatory in matrix and subordinate wh-questions withargument operators (compare a/a´) but impossible with the other operator types (b–d) in the dialectwith pattern 1.


a. Arg–V–S (Me pregunto) qué [se comió]V Miguel(wonder.1s) what ate.3s M

‘What did Miguel eat? / I wonder what Miguel ate.’a´. *Arg–S–V * (Me pregunto) qué Miguel [se comió]V

b. Loc–S–V (Me pregunto) dónde Miguel [se fue]V(wonder.1s) where M went.3s

‘Where did Miguel go? / I wonder where Miguel went.’c. Man–S–V (Me pregunto) cómo Miguel [se comportó]V

(wonder.1s) how M behaved.3s‘How did Miguel behave? / I wonder how Miguel behaved.’

d. Reas–S–V (Me pregunto) por qué Miguel [se enojó]V(wonder.1s) why M got-mad.3s

‘Why did Miguel get mad? / I wonder why Miguel got mad.’

Ranking STAY below ARGOP-SPEC but above the other OP-SPEC constraints as in (29b) explainsthe facts in this dialect: only argument operators can violate STAY twice to be in specifier position.OP-SCOPE demands that the other operators move too, but minimal violation of STAY will ensurethat the movement will be an adjunction and not a movement to specifier.

The constraint tableaux in (T1) show how decisions of optimality are made in this dialect.Because ARGOP-SPEC outranks STAY, inv wins in (T1a), but because STAY outranks the otherthree OP-SPEC constraints, adj wins in (T1b,c,d).

Eric J. Bakovic

21


a. Candidates: Argument operatorARGOP-

SPECSTAY

LOCOP-

SPEC

MANOP-

SPEC

REASOP-

SPEC

inv ☞ [VP qué [V´ se comió [VP Miguel [V´ tv twh ] ] ] ] **

adj [VP qué [VP Miguel [V´ se comió twh ] ] ] *! *

b. Candidates: Location operatorARGOP-

SPECSTAY

LOCOP-

SPEC

MANOP-

SPEC

REASOP-

SPEC

inv [VP dónde [V´ se fue [VP Miguel [V´ tv twh ] ] ] ] **!


c. Candidates: Manner operatorARGOP-

SPECSTAY

LOCOP-

SPEC

MANOP-

SPEC

REASOP-

SPEC



d. Candidates: Reason operatorARGOP-

SPECSTAY

LOCOP-

SPEC

MANOP-

SPEC

REASOP-

SPEC



2.2.3 Pattern 2: inversion with argument and location operators

As shown by the data in (D2), inversion is obligatory in matrix and subordinate wh-questions withargument and location operators (compare a/a´ and b/b´) but impossible with the other operatortypes (c–d) in the dialect with pattern 2.





‘Where did Miguel go? / I wonder where Miguel went.’b´. *Loc–S–V * (Me pregunto) dónde Miguel [se fue]V


22


‘How did Miguel behave? / I wonder how Miguel behaved.’d. Reas–S–V (Me pregunto) por qué Miguel [se enojó]V

(wonder.1s) why M got-mad.3s‘Why did Miguel get mad? / I wonder why Miguel got mad.’

Ranking STAY below LOCOP-SPEC (and necessarily below ARGOP-SPEC as a result) but abovethe other OP-SPEC constraints as in (29c) explains the facts in this dialect: only argument andlocation operators can violate STAY twice to be in specifier position.

The constraint tableaux in (T2) show how decisions of optimality are made in this dialect.Because ARGOP-SPEC and LOCOP-SPEC outrank STAY, inv wins in (T2a,b), but because STAYoutranks the other two OP-SPEC constraints, adj wins in (T2c,d).



SPEC

LOCOP-

SPECSTAY

MANOP-

SPEC

REASOP-

SPEC




SPEC

LOCOP-

SPECSTAY

MANOP-

SPEC

REASOP-

SPEC

inv ☞ [VP dónde [V´ se fue [VP Miguel [V´ tv twh ] ] ] ] **

adj [VP dónde [VP Miguel [V´ se fue twh ] ] ] *! *


SPEC

LOCOP-

SPECSTAY

MANOP-

SPEC

REASOP-

SPEC




SPEC

LOCOP-

SPECSTAY

MANOP-

SPEC

REASOP-

SPEC



Eric J. Bakovic

23

2.2.4 Pattern 3: inversion with argument, location and manner operators

As shown by the data in (D3), inversion is obligatory in matrix wh-questions with argument,location, and manner operators (compare a/a´, b/b´ and c/c´) but not with reason operators (d) inthe dialect with pattern 3.







‘How did Miguel behave? / I wonder how Miguel behaved.’c´. *Man–S–V * (Me pregunto) cómo Miguel [se comportó]V


‘Why did Miguel get mad? / I wonder why Miguel got mad.’

Ranking STAY below MANOP-SPEC (and necessarily below ARGOP-SPEC and LOCOP-SPEC as aresult) but above REASOP-SPEC as in (29d) explains the facts in this dialect: only argument,location and manner operators can violate STAY twice to be in specifier position.

The constraint tableaux in (T3) show how decisions of optimality are made in this dialect.Because STAY outranks REASOP-SPEC, adj wins in (T3d), but because the other three OP-SPECconstraints outrank STAY, inv wins in (T3a,b,c).



SPEC

LOCOP-

SPEC

MANOP-

SPECSTAY

REASOP-

SPEC




SPEC

LOCOP-

SPEC

MANOP-

SPECSTAY

REASOP-

SPEC




24


SPEC

LOCOP-

SPEC

MANOP-

SPECSTAY

REASOP-

SPEC

inv ☞ [VP cómo [V´ se comportó [VP Miguel [V´ tv twh ] ] ] ] **

adj [VP cómo [VP Miguel [V´ se comportó twh ] ] ] *! *


SPEC

LOCOP-

SPEC

MANOP-

SPECSTAY

REASOP-

SPEC



2.2.5 Pattern 4: inversion with all operators

As shown by the data in (D4), inversion is obligatory in matrix wh-questions with all operators inthe dialect with pattern 4.







‘How did Miguel behave? / I wonder how Miguel behaved.’c´. *Man–S–V * (Me pregunto) cómo Miguel [se comportó]V


‘Why did Miguel get mad? / I wonder why Miguel got mad.’d´. *Reas–S–V * (Me pregunto) por qué Miguel [se enojó]V

Ranking STAY below all of the OP-SPEC constraints as in (29e) explains the facts in this dialect:all operators can violate STAY twice to be in specifier position.

The constraint tableaux in (T4) show how decisions of optimality are made in this dialect.Because all of the OP-SPEC constraints outrank STAY, inv always wins.27

27 Note that this is the ranking that accounts for English matrix wh-questions where there is inversion with all operators,hence Grimshaw’s (1995) monolithic OP-SPEC.

Eric J. Bakovic

25



SPEC

LOCOP-

SPEC

MANOP-

SPEC

REASOP-

SPECSTAY




SPEC

LOCOP-

SPEC

MANOP-

SPEC

REASOP-

SPECSTAY




SPEC

LOCOP-

SPEC

MANOP-

SPEC

REASOP-

SPECSTAY

inv ☞ [VP cómo [V´ se comportó [VP Miguel [V´ tv twh ] ] ] ] **

adj [VP cómo [VP Miguel [V´ se comportó twh ] ] ] *! *


SPEC

LOCOP-

SPEC

MANOP-

SPEC

REASOP-

SPECSTAY

inv ☞ [VP por qué [V´ se enojó [VP Miguel [V´ tv twh ] ] ] ] **

adj [VP por qué [VP Miguel [V´ se enojó twh ] ] ] *! *

2.3 Concluding remarks

In this section I have provided an analysis of wh-movement and inversion in Spanish dialects withno matrix/subordinate distinction, explaining the limited variation in the obligatoriness of inversionby proposing a markedness subhierarchy of OP-SPEC constraints within which STAY can beranked anywhere. The fact that there is no other kind of variation is directly explained by themarkedness subhierarchy: Universal Grammar sets the relative ranking of the OP-SPECconstraints, and there is no way to disturb this ranking except by parochial ranking of a conflictingconstraint within the markedness subhierarchy, like STAY.

Factoring OB-HD out of the set of rankable/violable constraints has had two important,interdependent effects on the overall analysis in this section which are worth pointing out here.First, OB-HD isn’t among the constraints that can be ranked in different spots of the markednesssubhierarchy, and as such does not needlessly predict a larger typology of possible grammars.Second, and more directly relevant, the analysis relies crucially on the distinction between wh-movement and inversion and wh-adjunction. A dominating OP-SPEC constraint must force twoviolations of STAY rather than one — this crucially distinguishes OP-SCOPE, which only demandsthat operators have scope, from OP-SPEC, which prefers said scope to be taken from a specifier


26

position. This cannot be achieved with OB-HD thrown into the works of Con (the set ofconstraints provided by UG), as argued in §1.2.

Finally, it is worth noting here that I have been assuming all operators to be base-generatedwithin the lower V´, as the careful reader may have noted from the candidates considered in thetableaux. Absolutely nothing hinges on this assumption, however: there is no reference made byany of the constraints as to where the traces of the operators should be. There is only a constraintagainst the trace (STAY) and constraints on where the operators themselves should be (the OP-SPEC constraints). The analysis is compatible with the possibility that sentential adverbialoperators, like why, are base-generated clause-initially (as proposed by Rizzi 1990:51). Thecrucial STAY violation distinguishing inv from adj is the head-movement, not the wh-movement.If inv is to be the winner, why is base-generated in a clause-initial specifier position (forcing headmovement); if adj is to win, why is base-generated in a clause-initial adjoined position.28

3 The Matrix/Subordinate Distinction

In this section, I provide an analysis of the matrix/subordinate wh-question distinction in Spanishwith respect to inversion, which I argue involves the free ranking of another constraint (like STAY)in the rigid markedness subhierarchy. This constraint is part of Grimshaw’s (1995) PROJ-PRINconstraint, following the slight reworking in §1.2 of Grimshaw’s insightful analysis of the parallelmatrix/subordinate distinction in English.

3.1 Inversion in subordinate clauses

The same range of variation that exists in Spanish matrix wh-questions exists in subordinate wh-questions, but with a crucial generalization to be noted: there are no dialects that require inversionwith more operator types in subordinate wh-questions than in matrix wh-questions. This is thesubset relationship (3), repeated below in (30).

(30) Subset relationship between matrix and subordinate clause-types

In a given dialect, subordinate clauses display inversion with a subset of the wh-phrase types that matrix clauses display inversion with.

To make the point of (30) clearer, recall that what I have been calling patterns 0–4 up to thispoint have only been given those labels with respect to inversion in both matrix and subordinatewh-questions, when there is no matrix/subordinate distinction. I have been ignoring the fact thatdialects can be classified along both the matrix and subordinate dimension. If any combination ofthese variations were possible, we would have 5 x 5 = 25 different dialectal possibilities.According to (30), however, only fifteen of the possibilities are actualities. I repeat (4) as (31)below, showing the twenty-five different dialectal possibilities, with the actualities obeying (30)boxed. In addition, the dialects already accounted for in §2 (those without a matrix/subordinatedistinction) are shaded.29

28 Rizzi’s actual proposal was to base-generate the sentential adverbial operator in Spec/CP, but this is simply becauseSpec/CP is assumed to be the only place for operators. Given the freedom of choice between clause-initial specifier andadjoined position in the analysis proposed here, I allow myself the slight reinterpretation of his proposal seen in the text.29 The only varieties in (31) not attested in my survey are G, M and Y. Given the strength of the generalization, I interpretthis as a gap due to the small scale of the survey (30 native speakers). It may seem initially curious that these unattestedvarieties are all ones where with no matrix/subordinate distinction: 1/1, 2/2 and 4/4. However, it is hard to maintain thiscuriosity when one considers that 1/1 is Torrego’s (1984) dialect, and that the other two egalitarian varieties, 0/0 and 3/3,are strongly attested in my survey both in the number of speakers and in the consistency of their judgments.

Eric J. Bakovic

27

(31) Conceivable dialectal possibilities (actualities boxed; obeying (30))

Dialects A B C D E F G H I J K L M N O P Q R S T U V W X Y

Matrix 0 0 0 0 0 1 1 1 1 1 2 2 2 2 2 3 3 3 3 3 4 4 4 4 4

Subordinate 0 1 2 3 4 0 1 2 3 4 0 1 2 3 4 0 1 2 3 4 0 1 2 3 4

In the following subsections I show how the independent parochial ranking possibilities of PROJ-PRIN(H) and STAY in the markedness subhierarchy accounts for the subset relationship (30).

3.2 PROJ-PRIN(H) and STAY: the “stringency” relation30

As was demonstrated in §1.2, what the ranking of PROJ-PRIN(H) above OP-SPEC does in Englishis part of the work that STAY would do were it ranked above OP-SPEC — it demands wh-adjunction (no inversion), but in subordinate wh-questions only. OP-SPEC ranked above STAY, inturn, predicts inversion in matrix clauses. If PROJ-PRIN(H) does not dominate an OP-SPECconstraint that itself dominates STAY then, a distinction between matrix and subordinate wh-questions with respect to inversion is predicted not to exist.

As pointed out to me by Alan Prince (personal communication), this follows from the factthat STAY is more stringent than PROJ-PRIN(H). A violation of PROJ-PRIN(H) entails a violationof STAY but not vice versa, because movement into the head of a subordinate clause (what PROJ-PRIN(H) militates against) entails a trace (what STAY militates against) but not vice versa. Inaddition, STAY and PROJ-PRIN(H) do not directly conflict with each other. STAY demandseverything that PROJ-PRIN(H) demands, and more. This is what it means for two constraints to bein a stringency relation.

Prince notes that this quite common stringency relation is often confused with Prince &Smolensky’s (1993) Paninian relation, as defined in (32a). The definitions in (32b–d) aredefinitions relevant to the understanding of (32a) and of (33), a useful theorem based on thePaninian constraint relation.

(32) The Paninian constraint relation (and supporting definitions)

a. Dfn. Paninian relation. Let S and G be two constraints. S stands to G as specialto general in a Paninian relation if, for any input i to which S applies non-vacuously, any parse of i which satisfies S violates G.

b. Dfn. Non-vacuous application. A constraint C applies non-vacuously to an input iif C separates Gen(i), the set of candidate output parses of i admitted by UG.

c. Dfn. Separation. A constraint C separates a set of structures if C is satisfied bysome members of the set and violated by others.

d. Dfn. Activity. A constraint C is active on an input i in a constraint hierarchy CH ifC separates the set of candidate output parses of i which are admitted by the portionof CH which dominates C.

(33) Panini’s Theorem on constraint-ranking

Let S and G stand as specific to general in a Paninian constraint relation. SupposeS and G are part of a constraint hierarchy CH, and that G is active in CH on someinput i. Then if G » S, S is not active on i.

30 I thank Alan Prince and Paul Smolensky for helping me understand the details of the issues addressed in this subsection.


28

By Panini’s Theorem, the relative ranking of two constraints in a Paninian relation is rather simpleto determine: if the special constraint is active on some input, then it outranks the generalconstraint; if it is not active on any input, the general constraint outranks it. This seems similar tothe stringency relation, but as Prince points out in an unpublished proof, the similarity is illusory.

First and foremost, the definition of a Paninian relation between two constraints S and Grequires conflict between the constraints: “S stands to G as special to general in a Paninianrelation if, for any input i to which S applies non-vacuously, any parse of i which satisfies Sviolates G.” This isn’t the case between constraints related by stringency, as defined in (34).

(34) Stringency

a. Dfn. Stringency relation. Let S and G be two constraints. S stands to G as specialto general in a stringency relation if, for any input i to which S applies non-vacuously, any parse of i which violates S also violates G.

Secondly, “there is in fact no unqualified guarantee that … domination by the general constraint[e.g., STAY — E.J.B.] inactivates the more specific one [e.g., PROJ-PRIN(H) — E.J.B.].” Thisis a quote from Prince’s proof, which aims to “establish some conditions under which inactivation[of the special constraint under stringency — E.J.B.] can be guaranteed.” Here I want to establishone condition under which activation of the special constraint under stringency can be guaranteed— what I call independent activity. This condition is defined in (35) below. It is of course notmeant to be the only condition under which a special constraint under stringency can beindependently active; it is simply the one that has emerged in the present context.

(35) Condition on independent activity

Let S and G stand as specific to general in a stringency constraint relation. SupposeS and G are part of a constraint hierarchy CH, and that given an input i, S and Gboth separate Gen(i), the set of candidate outputs of i admitted by UG. S isindependently active on i if there is some constraint C such that (i) C separatesGen(i), (ii) C conflicts with both S and G on Gen(i) and (iii) S » C » G.

STAY is more stringent than PROJ-PRIN(H), and the OP-SPEC constraints conflict with both STAYand PROJ-PRIN(H). By the definition in (35), PROJ-PRIN(H) is independently active in thefollowing schema: PROJ-PRIN(H) » some OP-SPEC constraint » STAY.

The following simple demonstration with a monolithic OP-SPEC constraint makes thepoint. When either PROJ-PRIN(H) or STAY is ranked highest among the constraints OP-SPEC,STAY, and PROJ-PRIN(H), the result is wh-adjunction in subordinate wh-questions: violation ofOP-SPEC is preferred to a violation of PROJ-PRIN(H), or to the entailed second violation of STAY(the first being caused by wh-movement, always compelled by inviolable OP-SCOPE). This isshown in the tableaux in (36a–d). When OP-SPEC is ranked highest, the result is wh-movement tospecifier and inversion: the violation of PROJ-PRIN(H) and the concomitant second STAY violationare preferred to violation of OP-SPEC. This is shown in (36e,f).

(36) OP-SPEC, STAY, and PROJ-PRIN(H): subordinate wh-questions

a. P-P(H) OP-SPEC STAY b. P-P(H) STAY OP-SPEC

inv *! * * inv *! * *

adj ☞ * * adj ☞ * *

Eric J. Bakovic

29

c. STAY P-P(H) OP-SPEC d. STAY OP-SPEC P-P(H)

inv **! * inv **! *

adj ☞ * * adj ☞ * *

e. OP-SPEC P-P(H) STAY f. OP-SPEC STAY P-P(H)

inv ☞ * * * inv ☞ * * *

adj *! * adj *! *

Matrix and subordinate wh-questions are predicted to be identical with respect to inversion in(36b–f). Recall that PROJ-PRIN(H) is vacuously satisfied in matrix clauses. (36b,c,d) predict wh-adjunction in both matrix and subordinate wh-questions because STAY is ranked above OP-SPEC;(36e,f) predict wh-movement to specifier and inversion in both matrix and subordinate wh-questions because OP-SPEC dominates both STAY and PROJ-PRIN(H).

Only (36a) predicts a distinction between matrix and subordinate clauses. PROJ-PRIN(H),as we see in the tableau, chooses adj as optimal in subordinate clauses. Since PROJ-PRIN(H) isvacuously satisfied in matrix clauses, the next highest constraint, OP-SPEC, chooses inv as optimalin matrix clauses. This is the schema necessary for independent activity of PROJ-PRIN(H) by thecondition in (35): PROJ-PRIN(H) (= S) » OP-SPEC (= C) » STAY (= G).

So, even though there are thirty different ways to rank STAY and PROJ-PRIN(H) within themarkedness subhierarchy, only fifteen distinguishable systems emerge from these rankings,predicting exactly the boxed options in (31). Some of those rankings have one or more of the OP-SPEC constraints ranked between PROJ-PRIN(H) and STAY as in (36a), predicting amatrix/subordinate distinction; others do not meet the condition in (35) and as such predict nomatrix/subordinate distinction with respect to inversion.

3.3 Typological effects with the markedness subhierarchy 2: PROJ-PRIN(H) and STAY

The thirty different possible ways to rank all of the constraints, obeying the markednesssubhierarchy, are given below in (37). Shaded blocks of rankings are rankings that all make thesame grammatical predictions. Note that within each of these shaded blocks, PROJ-PRIN(H) isalways ranked either just above or somewhere below STAY, which is itself in the same spot of themarkedness subhierarchy. This configuration stifles the independent activity of PROJ-PRIN(H)with respect to these constraints, because it does not meet the condition in (35). Countingnonshaded rankings and shaded blocks of rankings as one each, there are fifteen distinguishablegrammars in (37).

(37) Thirty ranking permutations

1 PP RR OO JJ --PPRR II NN((HH)) » SS TT AA YY » ARGOP-SPEC » LOCOP-SPEC » MANOP-SPEC » REASOP-SPEC

2 SS TT AA YY » PP RR OO JJ --PPRR II NN((HH)) » ARGOP-SPEC » LOCOP-SPEC » MANOP-SPEC » REASOP-SPEC

3 SS TT AA YY » ARGOP-SPEC » PP RR OO JJ --PPRR II NN((HH)) » LOCOP-SPEC » MANOP-SPEC » REASOP-SPEC

4 SS TT AA YY » ARGOP-SPEC » LOCOP-SPEC » PP RR OO JJ --PPRR II NN((HH)) » MANOP-SPEC » REASOP-SPEC

5 SS TT AA YY » ARGOP-SPEC » LOCOP-SPEC » MANOP-SPEC » PP RR OO JJ --PPRR II NN((HH)) » REASOP-SPEC

6 SS TT AA YY » ARGOP-SPEC » LOCOP-SPEC » MANOP-SPEC » REASOP-SPEC » PP RR OO JJ --PPRR II NN (( HH ))


30

7 PP RR OO JJ --PPRR II NN((HH)) » ARGOP-SPEC » SS TT AA YY » LOCOP-SPEC » MANOP-SPEC » REASOP-SPEC

8 ARGOP-SPEC » PP RR OO JJ --PPRR II NN((HH)) » SS TT AA YY » LOCOP-SPEC » MANOP-SPEC » REASOP-SPEC

9 ARGOP-SPEC » SS TT AA YY » PP RR OO JJ --PPRR II NN((HH)) » LOCOP-SPEC » MANOP-SPEC » REASOP-SPEC

10 ARGOP-SPEC » SS TT AA YY » LOCOP-SPEC » PP RR OO JJ --PPRR II NN((HH)) » MANOP-SPEC » REASOP-SPEC

11 ARGOP-SPEC » SS TT AA YY » LOCOP-SPEC » MANOP-SPEC » PP RR OO JJ --PPRR II NN((HH)) » REASOP-SPEC

12 ARGOP-SPEC » SS TT AA YY » LOCOP-SPEC » MANOP-SPEC » REASOP-SPEC » PP RR OO JJ --PPRR II NN (( HH ))13 PP RR OO JJ --PPRR II NN((HH)) » ARGOP-SPEC » LOCOP-SPEC » SS TT AA YY » MANOP-SPEC » REASOP-SPEC

14 ARGOP-SPEC » PP RR OO JJ --PPRR II NN((HH)) » LOCOP-SPEC » SS TT AA YY » MANOP-SPEC » REASOP-SPEC

15 ARGOP-SPEC » LOCOP-SPEC » PP RR OO JJ --PPRR II NN((HH)) » SS TT AA YY » MANOP-SPEC » REASOP-SPEC

16 ARGOP-SPEC » LOCOP-SPEC » SS TT AA YY » PP RR OO JJ --PPRR II NN((HH)) » MANOP-SPEC » REASOP-SPEC

17 ARGOP-SPEC » LOCOP-SPEC » SS TT AA YY » MANOP-SPEC » PP RR OO JJ --PPRR II NN((HH)) » REASOP-SPEC

18 ARGOP-SPEC » LOCOP-SPEC » SS TT AA YY » MANOP-SPEC » REASOP-SPEC » PP RR OO JJ --PPRR II NN (( HH ))19 PP RR OO JJ --PPRR II NN((HH)) » ARGOP-SPEC » LOCOP-SPEC » MANOP-SPEC » SS TT AA YY » REASOP-SPEC

20 ARGOP-SPEC » PP RR OO JJ --PPRR II NN((HH)) » LOCOP-SPEC » MANOP-SPEC » SS TT AA YY » REASOP-SPEC

21 ARGOP-SPEC » LOCOP-SPEC » PP RR OO JJ --PPRR II NN((HH)) » MANOP-SPEC » SS TT AA YY » REASOP-SPEC

22 ARGOP-SPEC » LOCOP-SPEC » MANOP-SPEC » PP RR OO JJ --PPRR II NN((HH)) » SS TT AA YY » REASOP-SPEC

23 ARGOP-SPEC » LOCOP-SPEC » MANOP-SPEC » SS TT AA YY » PP RR OO JJ --PPRR II NN((HH)) » REASOP-SPEC

24 ARGOP-SPEC » LOCOP-SPEC » MANOP-SPEC » SS TT AA YY » REASOP-SPEC » PP RR OO JJ --PPRR II NN (( HH ))25 PP RR OO JJ --PPRR II NN ((HH)) » ARGOP-SPEC » LOCOP-SPEC » MANOP-SPEC » REASOP-SPEC » SS TT AA YY

26 ARGOP-SPEC » PP RR OO JJ --PPRR II NN ((HH)) » LOCOP-SPEC » MANOP-SPEC » REASOP-SPEC » SS TT AA YY

27 ARGOP-SPEC » LOCOP-SPEC » PP RR OO JJ --PPRR II NN ((HH)) » MANOP-SPEC » REASOP-SPEC » SS TT AA YY

28 ARGOP-SPEC » LOCOP-SPEC » MANOP-SPEC » PP RR OO JJ --PPRR II NN ((HH)) » REASOP-SPEC » SS TT AA YY

29 ARGOP-SPEC » LOCOP-SPEC » MANOP-SPEC » REASOP-SPEC» PP RR OO JJ --PPRR II NN ((HH)) » SS TT AA YY

30 ARGOP-SPEC » LOCOP-SPEC » MANOP-SPEC » REASOP-SPEC » SS TT AA YY » PP RR OO JJ --PPRR II NN (( HH ))

The following subsections explain how this is all so. Instead of going through each of the thirtyrankings, I have decided to concentrate on only six of them (rankings 13–18), all with STAYranked between LOCOP-SPEC and MANOP-SPEC, and considering the possible rerankings ofPROJ-PRIN(H) only. Since PROJ-PRIN(H) is irrelevant to matrix clauses, this stable ranking ofSTAY predicts pattern 2 in matrix clauses (henceforth, Matrix 2) — only argument and locationoperators require inversion (see §2.2.3 for details). Generalizing this demonstration ofsubordinate wh-question patterns with Matrix 2 to the other combinations of matrix andsubordinate wh-question patterns is a straightforward exercise left to the reader.

Keeping the ranking of STAY constant in this way, each of the different rankings of PROJ-PRIN(H) may or may not predict different patterns in subordinate clauses, as detailed under therankings in (38) below. PROJ-PRIN(H) only makes an independently active distinction betweenmatrix and subordinate wh-questions when it is ranked above an OP-SPEC constraint that is itselfranked above STAY, as in (38a) and (38b). Rankings (38c–f), all with PROJ-PRIN(H) rankedeither just above or somewhere below STAY, make the same grammatical prediction: no distinctionbetween matrix and subordinate wh-questions.

Eric J. Bakovic

31

(38) Typological effects with the markedness subhierarchy: Matrix 2, Subordinate …

a. PP--PP((HH)) » ARGOP-SPEC » LOCOP-SPEC » SSTTAAYY » MANOP-SPEC » REASOP-SPEC

⇒ Adjunction of all operators in subordinate clauses (Subordinate 0)

b. ARGOP-SPEC » PP--PP((HH)) » LOCOP-SPEC » SSTTAAYY » MANOP-SPEC » REASOP-SPEC

⇒ Inversion only with arguments in subordinate clauses (Subordinate 1)

c. ARGOP-SPEC » LOCOP-SPEC » PP--PP((HH)) » SSTTAAYY » MANOP-SPEC » REASOP-SPEC

d. ARGOP-SPEC » LOCOP-SPEC » SSTTAAYY » PP--PP((HH)) » MANOP-SPEC » REASOP-SPEC

e. ARGOP-SPEC » LOCOP-SPEC » SSTTAAYY » MANOP-SPEC » PP--PP((HH)) » REASOP-SPEC

f. ARGOP-SPEC » LOCOP-SPEC » SSTT AA YY » MANOP-SPEC » REASOP-SPEC » PP--PP((HH))⇒ Inversion only with arguments and locations in both clause-types (Subordinate 2)

In the same style as §2.2, the following subsections explain in detail how the predictionsare borne out for each case in (38). Since the relevant data can be read from the following tableauxand/or mixed-and-matched from that given in §2.2, I have not bothered to repeat it separately here.

3.3.1 Matrix 2, Subordinate 0: no inversion in subordinate clauses

All wh-movements without inversion in subordinate wh-questions in the dialect with Matrix 2,Subordinate 0 are grammatical. Ranking PROJ-PRIN(H) above all of the OP-SPEC constraints as in(38a) explains the facts in this dialect. No operator can violate PROJ-PRIN(H) just to be in specifierposition in this dialect. However, OP-SCOPE demands that the operator move. To satisfy OP-SCOPE, all it takes is adjunction of the operator, avoiding the PROJ-PRIN(H) violation (but ofcourse violating PROJ-PRIN(A); see §1.4.3). Any attempt to satisfy one of the OP-SPECconstraints by putting the operator in a specifier position fatally violates PROJ-PRIN(H).

The constraint tableaux in (T2/0) show how decisions of optimality are made forsubordinate clauses in this dialect. Because PROJ-PRIN(H) outranks all of the OP-SPECconstraints, adj always wins.

(T2/0) Matrix 2, Subordinate 0: Tableaux

a. Candidates: Subordinate clause, Argumentoperator

P-P(H)

ARGOP-

SPEC

LOCOP-

SPECSTAY

MANOP-

SPEC

REASOP-

SPEC

inv [VP qué [V´ se comió [VP Miguel [V´ tv twh ] ] ] ] *! **

adj ☞ [VP qué [VP Miguel [V´ se comió twh ] ] ] * *

b. Candidates: Subordinate clause, Locationoperator

P-P(H)

ARGOP-

SPEC

LOCOP-

SPECSTAY

MANOP-

SPEC

REASOP-

SPEC

inv [VP dónde [V´ se fue [VP Miguel [V´ tv twh ] ] ] ] *! **



32

c. Candidates: Subordinate clause, Manneroperator

P-P(H)

ARGOP-

SPEC

LOCOP-

SPECSTAY

MANOP-

SPEC

REASOP-

SPEC

inv [VP cómo [V´ se comportó [VP Miguel [V´ tv twh ] ] ] ] *! **


d. Candidates: Subordinate clause, Reasonoperator

P-P(H)

ARGOP-

SPEC

LOCOP-

SPECSTAY

MANOP-

SPEC

REASOP-

SPEC

inv [VP por qué [V´ se enojó [VP Miguel [V´ tv twh ] ] ] ] *! **


3.3.2 Matrix 2, Subordinate 1: inversion with arguments in subordinate clauses

Inversion is obligatory in subordinate wh-questions with argument operators but impossible withthe other operator types in the dialect with Matrix 2, Subordinate 1. Ranking PROJ-PRIN(H) belowARGOP-SPEC but above the other OP-SPEC constraints as in (38b) explains the facts in this dialect:only argument operators can violate PROJ-PRIN(H) to be in specifier position. OP-SCOPEdemands that the other operators move, but PROJ-PRIN(H) will ensure that the movement will bean adjunction and not a movement to specifier, as desired.

The constraint tableaux in (T2/1) show how decisions of optimality are made forsubordinate clauses in this dialect. Because ARGOP-SPEC outranks PROJ-PRIN(H), inv wins in(T2/1a), but because PROJ-PRIN(H) outranks the other three OP-SPEC constraints, adj wins in(T2/1b,c,d).



ARGOP-

SPEC

P-P(H)

LOCOP-

SPECSTAY

MANOP-

SPEC

REASOP-

SPEC

inv ☞ [VP qué [V´ se comió [VP Miguel [V´ tv twh ] ] ] ] * **



ARGOP-

SPEC

P-P(H)

LOCOP-

SPECSTAY

MANOP-

SPEC

REASOP-

SPEC

inv [VP dónde [V´ se fue [VP Miguel [V´ tv twh ] ] ] ] *! **


Eric J. Bakovic

33


ARGOP-

SPEC

P-P(H)

LOCOP-

SPECSTAY

MANOP-

SPEC

REASOP-

SPEC

inv [VP cómo [V´ se comportó [VP Miguel [V´ tv twh ] ] ] ] *! **



ARGOP-

SPEC

P-P(H)

LOCOP-

SPECSTAY

MANOP-

SPEC

REASOP-

SPEC

inv [VP por qué [V´ se enojó [VP Miguel [V´ tv twh ] ] ] ] *! **


3.3.3 Matrix 2, Subordinate 2: no matrix/subordinate distinction

There is no matrix/subordinate distinction in the dialect with Matrix 2, Subordinate 2. RankingPROJ-PRIN(H) just above or anywhere below STAY, as in (38c-f), will guarantee this result: fromany such point in the hierarchy, PROJ-PRIN(H) does not meet the condition in (35) needed for it tobe independently active with respect to the other constraints considered here, and as such cannotmake a decision between the candidates that the more general STAY would not make itself.

The constraint tableaux in (T2/2) show how decisions of optimality are made forsubordinate clauses in this dialect. PROJ-PRIN(H) and STAY are shown to be unranked withrespect to each other, but are both ranked together in the same spot of the markednesssubhierarchy. This emphasizes that fact that the ranking of PROJ-PRIN(H) simply doesn’t matteronce it is fails to meet the condition in (35); that is, once it is not ranked above STAY with anintervening OP-SPEC constraint. Because ARGOP-SPEC and LOCOP-SPEC outrank PROJ-PRIN(H)and STAY, inv wins in (T2/2a,b), but because STAY outranks the other two OP-SPEC constraints,adj wins in (T2/2c,d).



ARGOP-

SPEC

LOCOP-

SPEC

P-P(H) STAY

MANOP-

SPEC

REASOP-

SPEC

inv ☞ [VP qué [V´ se comió [VP Miguel [V´ tv twh ] ] ] ] * **



ARGOP-

SPEC

LOCOP-

SPEC

P-P(H) STAY

MANOP-

SPEC

REASOP-

SPEC

inv ☞ [VP dónde [V´ se fue [VP Miguel [V´ tv twh ] ] ] ] * **



34


ARGOP-

SPEC

LOCOP-

SPEC

P-P(H) STAY

MANOP-

SPEC

REASOP-

SPEC

inv [VP cómo [V´ se comportó [VP Miguel [V´ tv twh ] ] ] ] *! **!



ARGOP-

SPEC

LOCOP-

SPEC

P-P(H) STAY

MANOP-

SPEC

REASOP-

SPEC

inv [VP por qué [V´ se enojó [VP Miguel [V´ tv twh ] ] ] ] *! **!


Note that if PROJ-PRIN(H) is ranked immediately above STAY, it technically makes thedecision in (T2/2c,d). However, STAY makes exactly the same decision from its point in thehierarchy. This shows clearly that as long as PROJ-PRIN(H) is not ranked above an OP-SPECconstraint that is itself ranked above STAY, it isn’t independently active. More specifically, itshows that there can be no evidence (modulo other constraints that may crucially interact withthese) that would force a choice between the rankings in (32c-f), since they all have the sameconsequence — no matrix/subordinate distinction with respect to inversion.

3.4 Concluding remarks

This section has completed the analysis of wh-movement and inversion in Spanish matrix andsubordinate clauses. Given that there are five ways to display inversion (not at all, with argumentoperators, with argument and location operators, etc.) and given a distinction between matrix andsubordinate clauses, there are in principle twenty-five different dialectal possibilities. However,only fifteen of these obey the subset relationship (30): there are no dialects that require inversionwith more operator types in subordinate wh-questions than in matrix wh-questions.

The subset relationship itself follows from the fact that a constraint regulating onlysubordinate clauses, understood to be PROJ-PRIN(H) in this analysis, exists in the set ofconstraints provided by UG. The fifteen dialects follow from the interaction of the constraints:whenever PROJ-PRIN(H) is ranked just above or anywhere below STAY, its independent effectsare never felt, rendering many different rankings indistinguishable from one another.

Single rankings consistent with each of the fifteen distinguishable grammars are displayedfor easy reference and comparison in (39). The letters of the alphabet to their left correspond to theboxed dialectal possibilities (the dialectal actualities) of (31). All rankings where PROJ-PRIN(H)immediately dominates STAY or where STAY dominates PROJ-PRIN(H) are collapsed into rankingswhere STAY and PROJ-PRIN(H) are not ranked with respect to each other but are in the same spotof the markedness subhierarchy, since it was shown in this section that all of those rankings havethe same effect.

Eric J. Bakovic

35

(39) Fifteen distinguishable grammars

A {SS TT AA YY, PP RR OO JJ -- PP RR II NN (( HH ))} » ARGOP-SPEC » LOCOP-SPEC » MANOP-SPEC » REASOP-SPEC

F PP RR OO JJ -- PP RR II NN (( HH )) » ARGOP-SPEC » SS TT AA YY » LOCOP-SPEC » MANOP-SPEC » REASOP-SPEC

G ARGOP-SPEC » {SS TT AA YY, PP RR OO JJ -- PP RR II NN (( HH ))} » LOCOP-SPEC » MANOP-SPEC » REASOP-SPEC

K PP RR OO JJ -- PP RR II NN (( HH )) » ARGOP-SPEC » LOCOP-SPEC » SS TT AA YY » MANOP-SPEC » REASOP-SPEC

L ARGOP-SPEC » PP RR OO JJ -- PP RR II NN (( HH )) » LOCOP-SPEC » SS TT AA YY » MANOP-SPEC » REASOP-SPEC

M ARGOP-SPEC » LOCOP-SPEC » {SS TT AA YY, PP RR OO JJ -- PP RR II NN (( HH ))} » MANOP-SPEC » REASOP-SPEC

P PP RR OO JJ -- PP RR II NN (( HH )) » ARGOP-SPEC » LOCOP-SPEC » MANOP-SPEC » SS TT AA YY » REASOP-SPEC

Q ARGOP-SPEC » PP RR OO JJ -- PP RR II NN (( HH )) » LOCOP-SPEC » MANOP-SPEC » SS TT AA YY » REASOP-SPEC

R ARGOP-SPEC » LOCOP-SPEC » PP RR OO JJ -- PP RR II NN (( HH )) » MANOP-SPEC » SS TT AA YY » REASOP-SPEC

S ARGOP-SPEC » LOCOP-SPEC » MANOP-SPEC » {SS TT AA YY, PP RR OO JJ -- PP RR II NN (( HH ))} » REASOP-SPEC

U PP RR OO JJ -- PP RR II NN (( HH )) » ARGOP-SPEC » LOCOP-SPEC » MANOP-SPEC » REASOP-SPEC » SS TT AA YY

V ARGOP-SPEC » PP RR OO JJ -- PP RR II NN (( HH )) » LOCOP-SPEC » MANOP-SPEC » REASOP-SPEC » SS TT AA YY

W ARGOP-SPEC » LOCOP-SPEC » PP RR OO JJ -- PP RR II NN (( HH )) » MANOP-SPEC » REASOP-SPEC » SS TT AA YY

X ARGOP-SPEC » LOCOP-SPEC » MANOP-SPEC » PP RR OO JJ -- PP RR II NN (( HH )) » REASOP-SPEC » SS TT AA YY

Y ARGOP-SPEC » LOCOP-SPEC » MANOP-SPEC » REASOP-SPEC » {SS TT AA YY, PP RR OO JJ -- PP RR II NN (( HH ))}

This analysis shows clearly how a large number of constraint permutations distinguish a number ofactual grammars that is significantly smaller than the number of permutations. This follows fromthe only way that very general, conflicting constraints are allowed to interact in OT: throughdomination, and nothing else.

4 On Optionality and Learning31

As mentioned in §0, the OT analysis of inversion in this paper predicts that when adjunction (40b)is possible for a given wh-phrase in a given clause-type in a given dialect, inversion (40a) for thatwh-phrase in that clause-type in that dialect is impossible.

(40) Tree diagrams of inversion vs. no inversion

a. Inversion b. Adjunction (no inversion)VP

wh V´

V VP

Subj tv twh

VP

wh VP

Subj

V twh

V´

However, the particular word order exemplified by inversion, wh–verb–subject, is grammatical forevery wh-phrase in both clause-types in every dialect. Is this optionality, when wh–subject–verb

31 I thank Jane Grimshaw for encouraging me to pursue the issues addressed in this section.


36

word order (that exemplified by adjunction) is also grammatical? Suppose that it is. There arethree obvious ways to encode optionality in Optimality Theory, listed in (41).32

(41) Three ways to encode optionality in OT

a. Candidates in free variation incur the same constraint violations.b. Constraints distinguishing candidates in free variation are tied; i.e. not crucially

ranked with respect to each other.c. Speakers are in control of two (or more) total rankings, each candidate in free

variation being predicted by one of the rankings.

For obvious reasons, (41a) is not the right way to encode this supposed optionality. Inversionviolates STAY (and PROJ-PRIN(H), where applicable); adjunction violates one of the OP-SPECconstraints. If inversion and adjunction incurred the same constraint violations, there wouldalways be optionality! As Grimshaw noted at her MIT talk (see fn. 33), it is extremely difficult toengineer the candidates and constraints such that candidates in free variation violate the sameconstraints. Among other things, if one of those candidates in optimal in any other language ordialect, then so is/are the other candidate(s), by definition — if they violate the same constraints,they violate the same constraints, under any ranking.

Now consider the tied constraints idea in (41b). Suppose we have the following partialranking, where LOCOP-SPEC and STAY are unranked with respect to each other.

(42) Partial ranking with tied constraints

ARGOP-SPEC » {LOCOP-SPEC, STAY} » MANOP-SPEC » REASOP-SPEC

This ranking will predict that inversion and adjunction will be in free variation with locationoperators only; argument operators will require inversion and manner and reason operators willrequire adjunction. No dialect like this exists. This shows not only that this is not a good way toencode the present case of supposed optionality, but also that this may not be a good way toencode optionality in general: it predicts grammars that do not exist.

The idea in (41c) is out for similar reasons. If a speaker can be in control of more than onetotal ranking, then they could be in control of the two rankings in (43), which make the samepredictions as the ranking in (42) above.

(43) Two total rankings

a. ARGOP-SPEC » LOCOP-SPEC » STAY » MANOP-SPEC » REASOP-SPEC

b. ARGOP-SPEC » STAY » LOCOP-SPEC » MANOP-SPEC » REASOP-SPEC

This is not to say that bidialectism of this sort is strictly impossible — presumably, bilingualismhas to be captured in this way, and perhaps differences in register and other such things — but itcan’t be possible to account for optionality.

The only option left is to claim that this is a case of apparent optionality, in Grimshaw &Samek-Lodovici’s (1995) sense — when there is a choice between inversion word order andadjunction word order, they are each the optimal analysis of a different input. Adjunction, as Ihave argued, is the optimal analysis of the inputs discussed so far when STAY and/or PROJ-PRIN(H) dominates the relevant OP-SPEC constraint. Inversion is the optimal analysis of these

32 For those familiar with Grimshaw’s (handout from her) talk at the MIT Optimality in Syntax Workshop on May 20,1995, I have added (41c) to her list of ways to encode optionality in OT for completeness.

Eric J. Bakovic

37

inputs when the relevant OP-SPEC constraint dominates STAY and PROJ-PRIN(H). The sameword order as inversion must be the optimal analysis of another input.

Of course, we cannot simply declare an arbitrary difference between inputs to achieve thisgoal. Fortunately, this is not necessary — there is a clear difference between the orders wh–verb–subject and wh-subject–verb when they are both available. The former reflects the generalavailability in Spanish of verb–subject order, even in declaratives (see fn. 16), which is referred toin the generative literature as free inversion: “free” movement of the subject to the right of the verbphrase (see in particular Torrego 1984 and references cited there). As has been argued for Italianin Samek-Lodovici (1994, in preparation) and Grimshaw and Samek-Lodovici (1995), thismovement is not “free” but rather obligatory for unstressed, contrastively focused arguments — inthis case, subjects.33 This obligatoriness can be tested with the following diagnostics, amongothers (see the references cited above).

(44) Question–answer pairs

Q: ¿Quién se fue? Who left?A: Miguel. Miguel.

Se fue Miguel. left Miguel.*Miguel se fue. Miguel left.

(45) Association with solamente ‘only’

a. Solamente se fue Miguel. Only left Miguel.b. *Solamente Miguel se fue. Only Miguel left.

The first diagnostic (44) is a wh-question–answer pair, where the phrase in the answercorresponding to the wh-word (the subject) must be contrastively focused. Of course, the questioncan be answered simply with the phrase itself, as in the first answer. However, if the answer is acomplete sentence, then only one option is available — with the subject realized post-verbally.34

The second diagnostic (45) is focus association with the adverb solamente ‘only’, where thesubject must again be realized post-verbally to be associated with the adverb.

The exact same contrasts can be replicated when there is an option between wh–verb–subject and wh–subject–verb word orders in wh-questions. The subject–verb order simply cannotbe used in contexts where the subject is supposed to be contrastively focused. I use here examplesbased on my own judgments; these diagnostics have not received the benefit of a full-fledgedsurvey, but I have checked them with speakers with patterns different from my own.35

33 See Bonet (1990) for similar observations about Catalan. Bonet’s analysis differs from that of the references cited in thetext in that she assumes the postverbal, constrastive focus position of the subject to be its base-generated position, a right-branching Spec/VP, where the subject can optionally stay rather than move to the canonical Spec/IP position. Thisposition is then given a contrastive focus interpretation and special intonation by components separate from the syntax.34 The preverbal subject answer is possible only if the subject is stressed, another way to realize contrastive focus. Fordiscussion of this issue, see Grimshaw & Samek-Lodovici (1995); as they do, I treat the two ways of realizing contrastivefocus as separate phenomena. Since stressing is always a possible way to realize contrastive focus, I won’t point this outfor subsequent examples. An example marked as ungrammatical is so under regular intonation, with no subject stress.35 Since I speak the Matrix 4, Subordinate 3 dialect, I can only use subordinate wh-question examples with por qué ‘why’ —that’s the only place where I allow adjunction and hence also a distinct contrastively focused subject word order.


38

(46) Question–answer pairs

Q: ¿Quién es que te preguntas por qué se enojó?Who is it that you wonder why he got mad?

A: Miguel. Miguel.Me pregunto por qué se enojó Miguel. I wonder why got mad Miguel.*Me pregunto por qué Miguel se enojó. I wonder why Miguel got mad.

(47) Association with solamente ‘only’

a. Quiero saber por qué solamente se enojó Miguel.I want to know why only got mad Miguel.

b. *Quiero saber por qué solamente Miguel se enojó.I want to know why only Miguel got mad.

What these diagnostics show is that contrastive focus is by no means just an (obligatory)interpretation of a syntactic option, but rather an interpretive property of the input that motivates adeviation from the syntactic norm. The norm in Spanish is subject–verb word order, arising indeclaratives and in some wh-questions (depending on the speaker, the clause-type, and so forth).The deviation motivated by contrastively focused subjects is verb–subject word order. Recall from§1.4.1 that the input of a clause has an optional marking M of the arguments of the lexical head Las (contrastive) foci or topic-referring. Arguments marked as foci are subject to the followingconstraint (the subscript ƒ here is for “focus”).36

(48) Constraint on foci

ALIGNFOCUS: ALIGN (XPƒ, Left, YP, Right)Align the left edge of contrastively focused constituents with the right edge of amaximal projection. Failed by non-aligned foci.

From Grimshaw & Samek-Lodovici (1995)

When properly ranked in a constraint hierarchy, this constraint demands that an argument markedas contrastively focused be adjoined to a maximal projection. Details aside (see (Grimshaw and)Samek-Lodovici’s work), a contrastively focused subject must be right-adjoined to VP to satisfythis constraint. Assuming movement is required to achieve this goal, the resulting structure isshown schematically in (49) for a wh-question in Spanish. In (49a), we have a structure with acontrastively focused subject accompanied by wh-movement and inversion; in (49b), a structurewith a contrastively focused subject accompanied by wh-adjunction and no inversion.

36 The formulation of the constraint in (48) follows the generalized alignment schema proposed by McCarthy & Prince(1993). All that matters for our purposes is its structural consequence, shown below in (49).

Eric J. Bakovic

39

(49) Contrastively focused subject movement

a. with inversion b. with adjunctionVP

wh V´

V VP

VP Subjƒ

tsubj tv twh

VP

wh VP

VP Subjƒ

tsubj V twh

Both structures in (49) are string-wise identical, each for different reasons, to plain inversion (i.e.,without contrastively focused subject movement; compare (40a)). Plain inversion is precisely thestructure predicted not to be allowed by my analysis if adjunction is possible. We therefore havean explanation for the fact that this word order is always grammatical: it’s always availablethrough contrastively focused subject movement, whether inversion itself is obligatory or not.

Analysis of some of the similarities and differences between Italian and Spanish withrespect to contrastive focusing can be found in Samek-Lodovici (in preparation). A completeanalysis of this phenomenon in Spanish is unfortunately beyond the scope of this paper, and sothis explanation must await further investigation.

4.2 Why so many dialects?

Whether the explanation above is correct or not, the fact remains that the word order in question,that exemplified by inversion, is always a grammatical possibility. This fact may explain whythere are so many dialects of Spanish. I should warn the reader that the remarks to be made hereare entirely speculative, and may raise more questions than they attempt to answer.37

Stromswold (1990) notes in her study of the acquisition of the English auxiliary systemthat children learn auxiliary inversion in English matrix wh-questions in surprisingly distinctstages. First, there is a stage when they do not invert at all. Then, they begin to invert with whoand what. Where, when and how come next (the order of the individual stages here is not asclear), and finally inversion is acheived with why. Perhaps not surprisingly, this parallels theproposed markedness subhierarchy.

Suppose language acquisition proceeds through the demotion of constraints, as in thelearning algorithm proposed by Tesar & Smolensky (1993). Combining the inversion acquisitiondata with the constraint demotion idea, the acquisition of inversion can be seen as tentativedemotion of STAY down the markedness subhierarchy, tentative in the sense that it proceeds instages as if it were making sure not to go too far, possibly making errors.38 First, STAY will be atthe top of the hierarchy, and no inversion will be possible. Then, STAY will jump down a notch tojust below ARGOP-SPEC, allowing inversion only with who and what. In the end, if one isacquiring English anyway, STAY will end up at the bottom of the markedness subhierarchy.

37 The following emerged from what was already very speculative discussion of various issues with Paul Smolensky.However, he is not responsible for anything said here.38 I restrict myself here to matrix clauses, and so do not discuss PROJ-PRIN(H). Similar comments apply to this constraintas apply to STAY, of course.


40

Now consider the situation for any Spanish learner. Whether the dialect she is learningrequires inversion or not, she is hearing sentences with the word order of inversion: in some if notall cases, it is the result of contrastively focused subject movement. There seem to only be weakdiscourse-related cues telling the learner whether a given sentence has a contrastively focusedsubject or not (structural focus need not be accompanied by any special intonation or stress, paceBonet 1990). The learner can thus misinterpret contrastively focused subject movements asinversions or vice-versa — the former mistakenly triggering demotion of STAY, the lattermistakenly blocking demotion of STAY.

As a consequence of this ambiguous input, a learner’s entire input experience may violatethe subset relationship (30) or the markedness subhierarchy, but their grammar won’t —eventually, the learner arrives at a grammar (a ranking) that obeys both. This is because themarkedness subhierarchy is universally defined, and the subset relationship (30) is a consequenceof a stringency relationship between the universal constraints STAY and PROJ-PRIN(H) and all oftheir possible interactions within the markedness subhierarchy.

References

Ackema, Peter and Ad Neeleman. 1995. Optimal Questions. Holland Institute of Linguistics ms.Aoun, Joseph. 1986. Generalized Binding: The Syntax and Logical Form of Wh-interrogatives.

Studies in Generative Grammar 26. Dordrecht: Foris.Aoun, Joseph and Yen-hui Audrey Li. 1993a. Wh-Elements in Situ: Syntax or LF? Linguistic

Inquiry 24.3, 199-238.Aoun, Joseph and Yen-hui Audrey Li. 1993b. On Some Differences Between Chinese and

Japanese Wh-Elements. Linguistic Inquiry 24.3, 365-372.Bonet, Eulàlia. 1990. Subjects in Catalan. MIT Working Papers in Linguistics, 1-26.Cheng, Lisa. 1991. On the Typology of Wh-questions. MIT PhD disseration.Chomsky, Noam. 1982. Some Concepts and Consequences of the Theory of Government and

Binding. Linguistic Inquiry Monograph 6. Cambridge, MA: MIT Press.Chomsky, Noam. 1991. Some Notes on Economy of Derivation and Representation. In R.

Freidin (ed.), Principles and Parameters in Comparative Grammar, 417-454. Cambridge,MA: MIT Press.

Chomsky, Noam. 1993. A Minimalist Program for Linguistic Theory. In K. Hale and S. J.Keyser (eds.), The View from Building 20: Essays in Linguistics in Honor of SylvainBromberger. Cambridge, MA: MIT Press.

Cinque, Guglielmo. 1990. Types of A´-Dependencies. Linguistic Inquiry Monograph 17.Cambridge, MA: MIT Press.

Collins, Chris. 1994. Economy of Derivation and the Generalized Proper Binding Condition.Linguistic Inquiry 25, 45-61.

Déprez, Viviane. 1989. On the Typology of Syntactic Positions and the Nature of Chains: Moveα to the specifier of functional projections. MIT PhD dissertation.

Déprez, Viviane. 1991. WH-Movement: Adjunction and Substitution. In D. Bates (ed.), TheProceedings of the Tenth West Coast Conference on Formal Linguistics, 103-114.Stanford, CA: Center for the Study of Language and Information.

Epstein, Samuel David. 1992. Derivational Constraints on A´-Chain Formation. LinguisticInquiry 23, 235-259.

Eric J. Bakovic

41

Goodall, Grant. 1991a. On the Status of SPEC of IP. In D. Bates (ed.), The Proceedings of theTenth West Coast Conference on Formal Linguistics, 175-182. Stanford, CA: Center forthe Study of Language and Information.

Goodall, Grant. 1991b. Spec of IP and Spec of CP in Spanish Wh-Questions. Paper presentedat the Twenty-first Annual Linguistic Symposium on Romance Languages, University ofCalifornia, Santa Barbara.

Grimshaw, Jane. 1991. Extended Projection. Brandeis University ms.Grimshaw, Jane. 1995. Projection, Heads, and Optimality. Rutgers University ms. (To appear

in Linguistic Inquiry.)Grimshaw, Jane and Vieri Samek-Lodovici. 1995. Optimal Subjects. Rutgers University ms.

(To appear in Beckman, Urbanczyk, and Walsh (eds.), Papers in Optimality Theory,UMOP.)

Higginbotham, James and Robert May. 1981. Questions, Quantifiers and Crossing. TheLinguistic Review 1, 41-80.

Huang, C.-T. James. 1982. Logical Relations in Chinese and the Theory of Grammar. MIT PhDdissertation.

Jackendoff, Ray S. 1972. Semantic Interpretation in Generative Grammar. Current Studies inLinguistics 2. Cambridge, MA: MIT Press.

Legendre, Géraldine, Colin Wilson, Paul Smolensky, Kristin Homer and William Raymond.1995. Optimality and Wh-Extraction. Johns Hopkins University and University ofColorado ms. (To appear in Beckman, Urbanczyk, and Walsh (eds.), Papers in OptimalityTheory, UMOP.)

Li, Yen-hui Audrey. 1992. Indefinite Wh in Mandarin Chinese. Journal of East AsianLinguistics 1, 125–156.

McCarthy, John and Alan Prince. 1993. Generalized Alignment. University of Massachusetts,Amherst and Rutgers University ms.

McCloskey, James. 1992. Adjunction, Selection, and Embedded Verb Second. LinguisticsResearch Center Report LRC-92-07. University of California, Santa Cruz.

McDaniel, Dana. 1989. Partial and Multiple Wh-Movement. Natural Language and LinguisticTheory 7, 565-604.

Prince, Alan and Paul Smolensky. 1993. Optimality Theory: Constraint Interaction in GenerativeGrammar. RuCCS Technical Report #2. Piscataway, NJ: Rutgers University Center forCognitive Science. (To appear as a Linguistic Inquiry Monograph.)

Reis, Marga and Inger Rosengren. 1992. What do Wh-Imperatives tell us about Wh-Movement?Natural Language and Linguistic Theory 10, 79-118.

Rivero, María-Luisa. 1978. Topicalization and Wh Movement in Spanish. Linguistic Inquiry 9,513-517.

Rivero, María-Luisa. 1980. On Left-Dislocation and Topicalization in Spanish. LinguisticInquiry 11.2, 363-393.

Rizzi, Luigi. 1990a. Relativized Minimality. Linguistic Inquiry Monograph 16. Cambridge,MA: MIT Press.

Rizzi, Luigi. 1990b. Speculations on Verb Second. In J. Mascaró and M. Nespor (eds.), TheHenk van Riemsdijk Festschrift, Grammar in Progress, 375-386. Dordrecht: Foris.

Rizzi, Luigi. 1991. Residual Verb Second and the Wh Criterion. Technical Reports in Formaland Computational Linguistics 2. Geneva: Faculté des Lettres, Université de Genève.

Rizzi, Luigi and Ian Roberts. 1991. Complex Inversion in French. Probus 1.1.Samek-Lodovici, Vieri. 1994. Italian’s Focus Position. Rutgers University ms.


42

Samek-Lodovici, Vieri. In preparation. Topic and Focus Effects on Clause Structure: AnOptimality Theory Approach. Rutgers University PhD dissertation.

Smolensky, Paul. 1995. On the Internal Structure of the Constraint Component Con of UG.Handout from talk given at UCLA, April 7.

Stromswold, Karin. 1990. Learnability and the Acquisition of Auxiliaries. MassachusettsInstitute of Technology PhD dissertation.

Suñer, Margarita. 1991. Indirect Questions and the Structure of CP. In Campos and Martínez-Gil (eds.), Current Studies in Spanish Linguistics, 238-312. Washington, DC:Georgetown University Press

Suñer, Margarita. 1994. V-Movement and the Licensing of Argumental Wh-Phrases in Spanish.Natural Language and Linguistic Theory 12.2, 335-372.

Tesar, Bruce and Paul Smolensky. 1993. The Learnability of Optimality Theory: An Algorithmand Some Basic Complexity Results. Technical Report CU-CS-678-93. Department ofComputer Science, University of Colorado, Boulder. (Revised version to appear inLinguistic Inquiry.)

Torrego, Esther. 1984. On Inversion in Spanish and Some of Its Effects. Linguistic Inquiry15.1, 103-129.

Eric J. BakovicDepartment of LinguisticsRutgers University18 Seminary PlaceNew Brunswick, NJ [email protected]

A Markedness Subhierarchy in Syntax - ROAroa.rutgers.edu/files/80-0000/roa-80-bakovic-2.pdf · A...

Documents

Transcript of A Markedness Subhierarchy in Syntax - ROAroa.rutgers.edu/files/80-0000/roa-80-bakovic-2.pdf · A...